

Symbolic API

# MXNet.mx.SymbolicNode — Type.

SymbolicNode

SymbolicNode is the basic building block of the symbolic graph in MXNet.jl.

(self :: SymbolicNode)(args :: SymbolicNode...)
(self :: SymbolicNode)(; kwargs...)

Make a new node by composing self with args. Or the arguments can be specified using keyword arguments.

source

# Base.cos — Method.

cos(data)

Computes the element-wise cosine of the input array.

The input should be in radians (:math:2\pi rad equals 360 degrees).

.. math:: cos([0, \pi/4, \pi/2]) = [1, 0.707, 0]

The storage type of $cos$ output is always dense

Defined in src/operator/tensor/elemwise_unary_op_trig.cc:L63

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# Base.cosh — Method.

cosh(data)

Returns the hyperbolic cosine of the input array, computed element-wise.

.. math:: cosh(x) = 0.5\times(exp(x) + exp(-x))

The storage type of $cosh$ output is always dense

Defined in src/operator/tensor/elemwise_unary_op_trig.cc:L216

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# Base.reshape — Method.

reshape(sym::SymbolicNode, dim; reverse=false, name)
reshape(sym::SymbolicNode, dim...; reverse=false, name)

Reshape SymbolicNode operator

Some dimensions of the shape can take special values from the set {0, -1, -2, -3, -4}. The significance of each is explained below:

0 copy this dimension from the input to the output shape.

Example:
- input shape = (2,3,4), shape = (4,0,2), output shape = (4,3,2)
- input shape = (2,3,4), shape = (2,0,0), output shape = (2,3,4)
- -1 infers the dimension of the output shape by using the remainder of the input dimensions keeping the size of the new array same as that of the input array. At most one dimension of shape can be -1.
Example:
- input shape = (2,3,4), shape = (6,1,-1), output shape = (6,1,4)
- input shape = (2,3,4), shape = (3,-1,8), output shape = (3,1,8)
- input shape = (2,3,4), shape=(-1,), output shape = (24,)
- -2 copy all/remainder of the input dimensions to the output shape.
Example:
- input shape = (2,3,4), shape = (-2,), output shape = (2,3,4)
- input shape = (2,3,4), shape = (2,-2), output shape = (2,3,4)
- input shape = (2,3,4), shape = (-2,1,1), output shape = (2,3,4,1,1)
- -3 use the product of two consecutive dimensions of the input shape as the output dimension.
Example:
- input shape = (2,3,4), shape = (-3,4), output shape = (6,4)
- input shape = (2,3,4,5), shape = (-3,-3), output shape = (6,20)
- input shape = (2,3,4), shape = (0,-3), output shape = (2,12)
- input shape = (2,3,4), shape = (-3,-2), output shape = (6,4)
- -4 split one dimension of the input into two dimensions passed subsequent to -4 in shape (can contain -1).
Example:
- input shape = (2,3,4), shape = (-4,1,2,-2), output shape = (1,2,3,4)
- input shape = (2,3,4), shape = (2,-4,-1,3,-2), output shape = (2,1,3,4)

If the argument reverse is set to 1, then the special values are inferred from right to left.

Example:

with reverse=false, for input shape = (10,5,4), shape = (-1,0), output shape would be (40,5)
with reverse=true, output shape will be (50,4).

source

# Base.sin — Method.

sin(data)

Computes the element-wise sine of the input array.

The input should be in radians (:math:2\pi rad equals 360 degrees).

.. math:: sin([0, \pi/4, \pi/2]) = [0, 0.707, 1]

The storage type of $sin$ output depends upon the input storage type:

sin(default) = default
sin(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_trig.cc:L46

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# Base.sinh — Method.

sinh(data)

Returns the hyperbolic sine of the input array, computed element-wise.

.. math:: sinh(x) = 0.5\times(exp(x) - exp(-x))

The storage type of $sinh$ output depends upon the input storage type:

sinh(default) = default
sinh(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_trig.cc:L201

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# Base.tan — Method.

tan(data)

Computes the element-wise tangent of the input array.

The input should be in radians (:math:2\pi rad equals 360 degrees).

.. math:: tan([0, \pi/4, \pi/2]) = [0, 1, -inf]

The storage type of $tan$ output depends upon the input storage type:

tan(default) = default
tan(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_trig.cc:L83

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# Base.tanh — Method.

tanh(data)

Returns the hyperbolic tangent of the input array, computed element-wise.

.. math:: tanh(x) = sinh(x) / cosh(x)

The storage type of $tanh$ output depends upon the input storage type:

tanh(default) = default
tanh(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_trig.cc:L234

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.Variable — Method.

Variable(name :: Union{Symbol, AbstractString})

Create a symbolic variable with the given name. This is typically used as a placeholder. For example, the data node, acting as the starting point of a network architecture.

Arguments

Dict{Symbol, AbstractString} attrs: The attributes associated with this Variable.

source

# MXNet.mx.arccos — Method.

arccos(data)

Returns element-wise inverse cosine of the input array.

The input should be in range [-1, 1]. The output is in the closed interval :math:[0, \pi]

.. math:: arccos([-1, -.707, 0, .707, 1]) = [\pi, 3\pi/4, \pi/2, \pi/4, 0]

The storage type of $arccos$ output is always dense

Defined in src/operator/tensor/elemwise_unary_op_trig.cc:L123

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.arccosh — Method.

arccosh(data)

Returns the element-wise inverse hyperbolic cosine of the input array, computed element-wise.

The storage type of $arccosh$ output is always dense

Defined in src/operator/tensor/elemwise_unary_op_trig.cc:L264

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.arcsin — Method.

arcsin(data)

Returns element-wise inverse sine of the input array.

The input should be in the range [-1, 1]. The output is in the closed interval of [:math:-\pi/2, :math:\pi/2].

.. math:: arcsin([-1, -.707, 0, .707, 1]) = [-\pi/2, -\pi/4, 0, \pi/4, \pi/2]

The storage type of $arcsin$ output depends upon the input storage type:

arcsin(default) = default
arcsin(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_trig.cc:L104

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.arcsinh — Method.

arcsinh(data)

Returns the element-wise inverse hyperbolic sine of the input array, computed element-wise.

The storage type of $arcsinh$ output depends upon the input storage type:

arcsinh(default) = default
arcsinh(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_trig.cc:L250

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.arctan — Method.

arctan(data)

Returns element-wise inverse tangent of the input array.

The output is in the closed interval :math:[-\pi/2, \pi/2]

.. math:: arctan([-1, 0, 1]) = [-\pi/4, 0, \pi/4]

The storage type of $arctan$ output depends upon the input storage type:

arctan(default) = default
arctan(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_trig.cc:L144

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.arctanh — Method.

arctanh(data)

Returns the element-wise inverse hyperbolic tangent of the input array, computed element-wise.

The storage type of $arctanh$ output depends upon the input storage type:

arctanh(default) = default
arctanh(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_trig.cc:L281

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.broadcast_axes — Method.

broadcast_axes(data, axis, size)

broadcast_axes is an alias of broadcast_axis.

Broadcasts the input array over particular axes.

Broadcasting is allowed on axes with size 1, such as from (2,1,3,1) to (2,8,3,9). Elements will be duplicated on the broadcasted axes.

Example::

// given x of shape (1,2,1) x = [[[ 1.], [ 2.]]]

// broadcast x on on axis 2 broadcast_axis(x, axis=2, size=3) = [[[ 1., 1., 1.], [ 2., 2., 2.]]] // broadcast x on on axes 0 and 2 broadcast_axis(x, axis=(0,2), size=(2,3)) = [[[ 1., 1., 1.], [ 2., 2., 2.]], [[ 1., 1., 1.], [ 2., 2., 2.]]]

Defined in src/operator/tensor/broadcast_reduce_op_value.cc:L207

Arguments

data::NDArray-or-SymbolicNode: The input
axis::Shape(tuple), optional, default=[]: The axes to perform the broadcasting.
size::Shape(tuple), optional, default=[]: Target sizes of the broadcasting axes.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.broadcast_axis — Method.

broadcast_axis(data, axis, size)

Broadcasts the input array over particular axes.

Broadcasting is allowed on axes with size 1, such as from (2,1,3,1) to (2,8,3,9). Elements will be duplicated on the broadcasted axes.

Example::

// given x of shape (1,2,1) x = [[[ 1.], [ 2.]]]

// broadcast x on on axis 2 broadcast_axis(x, axis=2, size=3) = [[[ 1., 1., 1.], [ 2., 2., 2.]]] // broadcast x on on axes 0 and 2 broadcast_axis(x, axis=(0,2), size=(2,3)) = [[[ 1., 1., 1.], [ 2., 2., 2.]], [[ 1., 1., 1.], [ 2., 2., 2.]]]

Defined in src/operator/tensor/broadcast_reduce_op_value.cc:L207

Arguments

data::NDArray-or-SymbolicNode: The input
axis::Shape(tuple), optional, default=[]: The axes to perform the broadcasting.
size::Shape(tuple), optional, default=[]: Target sizes of the broadcasting axes.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.broadcast_to — Method.

broadcast_to(data, shape)

Broadcasts the input array to a new shape.

Broadcasting is a mechanism that allows NDArrays to perform arithmetic operations with arrays of different shapes efficiently without creating multiple copies of arrays. Also see, Broadcasting <https://docs.scipy.org/doc/numpy/user/basics.broadcasting.html>_ for more explanation.

Broadcasting is allowed on axes with size 1, such as from (2,1,3,1) to (2,8,3,9). Elements will be duplicated on the broadcasted axes.

For example::

broadcast_to([[1,2,3]], shape=(2,3)) = [[ 1., 2., 3.], [ 1., 2., 3.]])

The dimension which you do not want to change can also be kept as 0 which means copy the original value. So with shape=(2,0), we will obtain the same result as in the above example.

Defined in src/operator/tensor/broadcast_reduce_op_value.cc:L231

Arguments

data::NDArray-or-SymbolicNode: The input
shape::Shape(tuple), optional, default=[]: The shape of the desired array. We can set the dim to zero if it's same as the original. E.g A = broadcast_to(B, shape=(10, 0, 0)) has the same meaning as A = broadcast_axis(B, axis=0, size=10).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.clip — Method.

clip(data, a_min, a_max)

Clips (limits) the values in an array.

Given an interval, values outside the interval are clipped to the interval edges. Clipping $x$ between a_min and a_x would be::

clip(x, a_min, a_max) = max(min(x, a_max), a_min))

Example::

x = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]

clip(x,1,8) = [ 1.,  1.,  2.,  3.,  4.,  5.,  6.,  7.,  8.,  8.]

The storage type of $clip$ output depends on storage types of inputs and the a_min, a_max parameter values:

clip(default) = default
clip(row_sparse, a_min <= 0, a_max >= 0) = row_sparse
clip(csr, a_min <= 0, a_max >= 0) = csr
clip(row_sparse, a_min < 0, a_max < 0) = default
clip(row_sparse, a_min > 0, a_max > 0) = default
clip(csr, a_min < 0, a_max < 0) = csr
clip(csr, a_min > 0, a_max > 0) = csr

Defined in src/operator/tensor/matrix_op.cc:L490

Arguments

data::NDArray-or-SymbolicNode: Input array.
a_min::float, required: Minimum value
a_max::float, required: Maximum value
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.expand_dims — Method.

expand_dims(data, axis)

Inserts a new axis of size 1 into the array shape

For example, given $x$ with shape $(2,3,4)$, then $expand_dims(x, axis=1)$ will return a new array with shape $(2,1,3,4)$.

Defined in src/operator/tensor/matrix_op.cc:L293

Arguments

data::NDArray-or-SymbolicNode: Source input
axis::int, required: Position where new axis is to be inserted. Suppose that the input NDArray's dimension is ndim, the range of the inserted axis is [-ndim, ndim]
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.log_softmax — Method.

log_softmax(data, axis)

Computes the log softmax of the input. This is equivalent to computing softmax followed by log.

Examples::

x = mx.nd.array([1, 2, .1]) mx.nd.log_softmax(x).asnumpy()

array([-1.41702998, -0.41702995, -2.31702995], dtype=float32)

x = mx.nd.array( [[1, 2, .1],[.1, 2, 1]] ) mx.nd.log_softmax(x, axis=0).asnumpy()

array([[-0.34115392, -0.69314718, -1.24115396], [-1.24115396, -0.69314718, -0.34115392]], dtype=float32)

Arguments

data::NDArray-or-SymbolicNode: The input array.
axis::int, optional, default='-1': The axis along which to compute softmax.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.relu — Method.

relu(data)

Computes rectified linear.

.. math:: max(features, 0)

The storage type of $relu$ output depends upon the input storage type:

relu(default) = default
relu(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L83

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.sigmoid — Method.

sigmoid(data)

Computes sigmoid of x element-wise.

.. math:: y = 1 / (1 + exp(-x))

The storage type of $sigmoid$ output is always dense

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L102

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.softmax — Method.

softmax(data, axis)

Applies the softmax function.

The resulting array contains elements in the range (0,1) and the elements along the given axis sum up to 1.

.. math:: softmax(\mathbf{z})j = \frac{e^{z_j}}{\sum^K e^{z_k}}

for :math:j = 1, ..., K

Example::

x = [[ 1. 1. 1.] [ 1. 1. 1.]]

softmax(x,axis=0) = [[ 0.5 0.5 0.5] [ 0.5 0.5 0.5]]

softmax(x,axis=1) = [[ 0.33333334, 0.33333334, 0.33333334], [ 0.33333334, 0.33333334, 0.33333334]]

Defined in src/operator/nn/softmax.cc:L54

Arguments

data::NDArray-or-SymbolicNode: The input array.
axis::int, optional, default='-1': The axis along which to compute softmax.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.@var — Macro.

@var <symbols>...

A handy macro for creating mx.Variable.

julia> x = @mx.var x
MXNet.mx.SymbolicNode x

julia> x, y, z = @mx.var x y z
(MXNet.mx.SymbolicNode x, MXNet.mx.SymbolicNode y, MXNet.mx.SymbolicNode z)

source

# Base.Iterators.Flatten — Method.

Flatten(data)

Flattens the input array into a 2-D array by collapsing the higher dimensions.

.. note:: Flatten is deprecated. Use flatten instead.

For an input array with shape $(d1, d2, ..., dk)$, flatten operation reshapes the input array into an output array of shape $(d1, d2...dk)$.

Note that the bahavior of this function is different from numpy.ndarray.flatten, which behaves similar to mxnet.ndarray.reshape((-1,)).

Example::

x = [[
    [1,2,3],
    [4,5,6],
    [7,8,9]
],
[    [1,2,3],
    [4,5,6],
    [7,8,9]
]],

flatten(x) = [[ 1.,  2.,  3.,  4.,  5.,  6.,  7.,  8.,  9.],
   [ 1.,  2.,  3.,  4.,  5.,  6.,  7.,  8.,  9.]]

Defined in src/operator/tensor/matrix_op.cc:L212

Arguments

data::NDArray-or-SymbolicNode: Input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# Base.:* — Method.

.*(x, y)

Elementwise multiplication of SymbolicNode.

source

# Base.:+ — Method.

+(args...)
.+(args...)

Elementwise summation of SymbolicNode.

source

# Base.:- — Method.

-(x, y)
.-(x, y)

Elementwise substraction of SymbolicNode. Operating with Real is available.

source

# Base.:/ — Method.

./(x, y)

Elementwise dividing a SymbolicNode by a scalar or another SymbolicNode

of the same shape.

Elementwise divide a scalar by an SymbolicNode.
Matrix division (solving linear systems) is not implemented yet.

source

# Base.:^ — Function.

.^(x, y)

Elementwise power of SymbolicNode and NDArray. Operating with Real is available.

source

# Base.LinAlg.dot — Method.

dot(lhs, rhs, transpose_a, transpose_b)

Dot product of two arrays.

$dot$'s behavior depends on the input array dimensions:

1-D arrays: inner product of vectors
2-D arrays: matrix multiplication
N-D arrays: a sum product over the last axis of the first input and the first axis of the second input

For example, given 3-D $x$ with shape (n,m,k) and $y$ with shape (k,r,s), the result array will have shape (n,m,r,s). It is computed by::

dot(x,y)[i,j,a,b] = sum(x[i,j,:]*y[:,a,b])

Example::

x = reshape([0,1,2,3,4,5,6,7], shape=(2,2,2)) y = reshape([7,6,5,4,3,2,1,0], shape=(2,2,2)) dot(x,y)[0,0,1,1] = 0 sum(x[0,0,:]*y[:,1,1]) = 0

The storage type of $dot$ output depends on storage types of inputs and transpose options:

dot(csr, default) = default
dot(csr.T, default) = row_sparse
dot(csr, row_sparse) = default
dot(default, csr) = csr
otherwise, $dot$ generates output with default storage

Defined in src/operator/tensor/dot.cc:L62

Arguments

lhs::NDArray-or-SymbolicNode: The first input
rhs::NDArray-or-SymbolicNode: The second input
transpose_a::boolean, optional, default=0: If true then transpose the first input before dot.
transpose_b::boolean, optional, default=0: If true then transpose the second input before dot.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# Base.LinAlg.norm — Method.

norm(data)

Flattens the input array and then computes the l2 norm.

Examples::

x = [[1, 2], [3, 4]]

norm(x) = [5.47722578]

rsp = x.cast_storage('row_sparse')

norm(rsp) = [5.47722578]

csr = x.cast_storage('csr')

norm(csr) = [5.47722578]

Defined in src/operator/tensor/broadcast_reduce_op_value.cc:L266

Arguments

data::NDArray-or-SymbolicNode: Source input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# Base.Math.cbrt — Method.

cbrt(data)

Returns element-wise cube-root value of the input.

.. math:: cbrt(x) = \sqrt[3]{x}

Example::

cbrt([1, 8, -125]) = [1, 2, -5]

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L601

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# Base.Math.gamma — Method.

gamma(data)

Returns the gamma function (extension of the factorial function to the reals), computed element-wise on the input array.

The storage type of $gamma$ output is always dense

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# Base._div — Method.

_div(lhs, rhs)

_div is an alias of elemwise_div.

Divides arguments element-wise.

The storage type of $elemwise_div$ output is always dense

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# Base._sub — Method.

_sub(lhs, rhs)

_sub is an alias of elemwise_sub.

Subtracts arguments element-wise.

The storage type of $elemwise_sub$ output depends on storage types of inputs

elemwise_sub(row_sparse, row_sparse) = row_sparse
elemwise_sub(csr, csr) = csr
otherwise, $elemwise_sub$ generates output with default storage

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# Base.abs — Method.

abs(data)

Returns element-wise absolute value of the input.

Example::

abs([-2, 0, 3]) = [2, 0, 3]

The storage type of $abs$ output depends upon the input storage type:

abs(default) = default
abs(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L385

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# Base.ceil — Method.

ceil(data)

Returns element-wise ceiling of the input.

The ceil of the scalar x is the smallest integer i, such that i >= x.

Example::

ceil([-2.1, -1.9, 1.5, 1.9, 2.1]) = [-2., -1., 2., 2., 3.]

The storage type of $ceil$ output depends upon the input storage type:

ceil(default) = default
ceil(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L463

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# Base.copy — Method.

copy(self :: SymbolicNode)

Make a copy of a SymbolicNode. The same as making a deep copy.

source

# Base.deepcopy — Method.

deepcopy(self :: SymbolicNode)

Make a deep copy of a SymbolicNode.

source

# Base.exp — Method.

exp(data)

Returns element-wise exponential value of the input.

.. math:: exp(x) = e^x \approx 2.718^x

Example::

exp([0, 1, 2]) = [1., 2.71828175, 7.38905621]

The storage type of $exp$ output is always dense

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L641

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# Base.expm1 — Method.

expm1(data)

Returns $exp(x) - 1$ computed element-wise on the input.

This function provides greater precision than $exp(x) - 1$ for small values of $x$.

The storage type of $expm1$ output depends upon the input storage type:

expm1(default) = default
expm1(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L720

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# Base.floor — Method.

floor(data)

Returns element-wise floor of the input.

The floor of the scalar x is the largest integer i, such that i <= x.

Example::

floor([-2.1, -1.9, 1.5, 1.9, 2.1]) = [-3., -2., 1., 1., 2.]

The storage type of $floor$ output depends upon the input storage type:

floor(default) = default
floor(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L482

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# Base.getindex — Method.

getindex(self :: SymbolicNode, idx :: Union{Int, Base.Symbol, AbstractString})

Get a node representing the specified output of this node. The index could be a symbol or string indicating the name of the output, or a 1-based integer indicating the index, as in the list of list_outputs.

source

# Base.identity — Method.

identity(data)

identity is an alias of _copy.

Returns a copy of the input.

From:src/operator/tensor/elemwise_unary_op_basic.cc:111

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# Base.log — Method.

log(data)

Returns element-wise Natural logarithmic value of the input.

The natural logarithm is logarithm in base e, so that $log(exp(x)) = x$

The storage type of $log$ output is always dense

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L653

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# Base.log10 — Method.

log10(data)

Returns element-wise Base-10 logarithmic value of the input.

$10**log10(x) = x$

The storage type of $log10$ output is always dense

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L665

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# Base.log1p — Method.

log1p(data)

Returns element-wise $log(1 + x)$ value of the input.

This function is more accurate than $log(1 + x)$ for small $x$ so that :math:1+x\approx 1

The storage type of $log1p$ output depends upon the input storage type:

log1p(default) = default
log1p(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L702

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# Base.log2 — Method.

log2(data)

Returns element-wise Base-2 logarithmic value of the input.

$2**log2(x) = x$

The storage type of $log2$ output is always dense

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L677

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# Base.max — Method.

max(data, axis, keepdims, exclude)

Computes the max of array elements over given axes.

Defined in src/operator/tensor/broadcast_reduce_op_value.cc:L160

Arguments

data::NDArray-or-SymbolicNode: The input
axis::Shape(tuple), optional, default=[]: The axis or axes along which to perform the reduction.

`` The default,axis=(), will compute over all elements into a scalar array with shape(1,)`.

If axis is int, a reduction is performed on a particular axis.

If axis is a tuple of ints, a reduction is performed on all the axes specified in the tuple.

If exclude is true, reduction will be performed on the axes that are NOT in axis instead.

Negative values means indexing from right to left. `` *keepdims::boolean, optional, default=0: If this is set toTrue, the reduced axes are left in the result as dimension with size one. *exclude::boolean, optional, default=0: Whether to perform reduction on axis that are NOT in axis instead. *name::Symbol: The name of theSymbolicNode. (e.g.:my_symbol), optional. *attrs::Dict{Symbol, AbstractString}: The attributes associated with thisSymbolicNode`.

source

# Base.mean — Method.

mean(data, axis, keepdims, exclude)

Computes the mean of array elements over given axes.

Defined in src/operator/tensor/broadcast_reduce_op_value.cc:L101

Arguments

data::NDArray-or-SymbolicNode: The input
axis::Shape(tuple), optional, default=[]: The axis or axes along which to perform the reduction.

`` The default,axis=(), will compute over all elements into a scalar array with shape(1,)`.

If axis is int, a reduction is performed on a particular axis.

If axis is a tuple of ints, a reduction is performed on all the axes specified in the tuple.

If exclude is true, reduction will be performed on the axes that are NOT in axis instead.

Negative values means indexing from right to left. `` *keepdims::boolean, optional, default=0: If this is set toTrue, the reduced axes are left in the result as dimension with size one. *exclude::boolean, optional, default=0: Whether to perform reduction on axis that are NOT in axis instead. *name::Symbol: The name of theSymbolicNode. (e.g.:my_symbol), optional. *attrs::Dict{Symbol, AbstractString}: The attributes associated with thisSymbolicNode`.

source

# Base.min — Method.

min(data, axis, keepdims, exclude)

Computes the min of array elements over given axes.

Defined in src/operator/tensor/broadcast_reduce_op_value.cc:L174

Arguments

data::NDArray-or-SymbolicNode: The input
axis::Shape(tuple), optional, default=[]: The axis or axes along which to perform the reduction.

`` The default,axis=(), will compute over all elements into a scalar array with shape(1,)`.

If axis is int, a reduction is performed on a particular axis.

If axis is a tuple of ints, a reduction is performed on all the axes specified in the tuple.

If exclude is true, reduction will be performed on the axes that are NOT in axis instead.

Negative values means indexing from right to left. `` *keepdims::boolean, optional, default=0: If this is set toTrue, the reduced axes are left in the result as dimension with size one. *exclude::boolean, optional, default=0: Whether to perform reduction on axis that are NOT in axis instead. *name::Symbol: The name of theSymbolicNode. (e.g.:my_symbol), optional. *attrs::Dict{Symbol, AbstractString}: The attributes associated with thisSymbolicNode`.

source

# Base.print — Function.

print([io::IO], sym::SymbolicNode)

Print the content of symbol, used for debug.

julia> layer = @mx.chain mx.Variable(:data)           =>
         mx.FullyConnected(name=:fc1, num_hidden=128) =>
         mx.Activation(name=:relu1, act_type=:relu)
MXNet.mx.SymbolicNode(MXNet.mx.MX_SymbolHandle(Ptr{Void} @0x000055b29b9c3520))

julia> print(layer)
Symbol Outputs:
        output[0]=relu1(0)
Variable:data
Variable:fc1_weight
Variable:fc1_bias
--------------------
Op:FullyConnected, Name=fc1
Inputs:
        arg[0]=data(0) version=0
        arg[1]=fc1_weight(0) version=0
        arg[2]=fc1_bias(0) version=0
Attrs:
        num_hidden=128
--------------------
Op:Activation, Name=relu1
Inputs:
        arg[0]=fc1(0)
Attrs:
        act_type=relu

source

# Base.prod — Method.

prod(data, axis, keepdims, exclude)

Computes the product of array elements over given axes.

Defined in src/operator/tensor/broadcast_reduce_op_value.cc:L116

Arguments

data::NDArray-or-SymbolicNode: The input
axis::Shape(tuple), optional, default=[]: The axis or axes along which to perform the reduction.

`` The default,axis=(), will compute over all elements into a scalar array with shape(1,)`.

If axis is int, a reduction is performed on a particular axis.

If axis is a tuple of ints, a reduction is performed on all the axes specified in the tuple.

If exclude is true, reduction will be performed on the axes that are NOT in axis instead.

Negative values means indexing from right to left. `` *keepdims::boolean, optional, default=0: If this is set toTrue, the reduced axes are left in the result as dimension with size one. *exclude::boolean, optional, default=0: Whether to perform reduction on axis that are NOT in axis instead. *name::Symbol: The name of theSymbolicNode. (e.g.:my_symbol), optional. *attrs::Dict{Symbol, AbstractString}: The attributes associated with thisSymbolicNode`.

source

# Base.repeat — Method.

repeat(data, repeats, axis)

Repeats elements of an array.

By default, $repeat$ flattens the input array into 1-D and then repeats the elements::

x = [[ 1, 2], [ 3, 4]]

repeat(x, repeats=2) = [ 1., 1., 2., 2., 3., 3., 4., 4.]

The parameter $axis$ specifies the axis along which to perform repeat::

repeat(x, repeats=2, axis=1) = [[ 1., 1., 2., 2.], [ 3., 3., 4., 4.]]

repeat(x, repeats=2, axis=0) = [[ 1., 2.], [ 1., 2.], [ 3., 4.], [ 3., 4.]]

repeat(x, repeats=2, axis=-1) = [[ 1., 1., 2., 2.], [ 3., 3., 4., 4.]]

Defined in src/operator/tensor/matrix_op.cc:L563

Arguments

data::NDArray-or-SymbolicNode: Input data array
repeats::int, required: The number of repetitions for each element.
axis::int or None, optional, default='None': The axis along which to repeat values. The negative numbers are interpreted counting from the backward. By default, use the flattened input array, and return a flat output array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# Base.reverse — Method.

reverse(data, axis)

Reverses the order of elements along given axis while preserving array shape.

Note: reverse and flip are equivalent. We use reverse in the following examples.

Examples::

x = [[ 0., 1., 2., 3., 4.], [ 5., 6., 7., 8., 9.]]

reverse(x, axis=0) = [[ 5., 6., 7., 8., 9.], [ 0., 1., 2., 3., 4.]]

reverse(x, axis=1) = [[ 4., 3., 2., 1., 0.], [ 9., 8., 7., 6., 5.]]

Defined in src/operator/tensor/matrix_op.cc:L665

Arguments

data::NDArray-or-SymbolicNode: Input data array
axis::Shape(tuple), required: The axis which to reverse elements.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# Base.round — Method.

round(data)

Returns element-wise rounded value to the nearest integer of the input.

Example::

round([-1.5, 1.5, -1.9, 1.9, 2.1]) = [-2., 2., -2., 2., 2.]

The storage type of $round$ output depends upon the input storage type:

round(default) = default
round(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L423

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# Base.sign — Method.

sign(data)

Returns element-wise sign of the input.

Example::

sign([-2, 0, 3]) = [-1, 0, 1]

The storage type of $sign$ output depends upon the input storage type:

sign(default) = default
sign(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L404

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# Base.sort — Method.

sort(data, axis, is_ascend)

Returns a sorted copy of an input array along the given axis.

Examples::

x = [[ 1, 4], [ 3, 1]]

// sorts along the last axis sort(x) = [[ 1., 4.], [ 1., 3.]]

// flattens and then sorts sort(x) = [ 1., 1., 3., 4.]

// sorts along the first axis sort(x, axis=0) = [[ 1., 1.], [ 3., 4.]]

// in a descend order sort(x, is_ascend=0) = [[ 4., 1.], [ 3., 1.]]

Defined in src/operator/tensor/ordering_op.cc:L126

Arguments

data::NDArray-or-SymbolicNode: The input array
axis::int or None, optional, default='-1': Axis along which to choose sort the input tensor. If not given, the flattened array is used. Default is -1.
is_ascend::boolean, optional, default=1: Whether to sort in ascending or descending order.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# Base.split — Method.

split(data, num_outputs, axis, squeeze_axis)

split is an alias of SliceChannel.

Splits an array along a particular axis into multiple sub-arrays.

.. note:: $SliceChannel$ is deprecated. Use $split$ instead.

Note that num_outputs should evenly divide the length of the axis along which to split the array.

Example::

x = [[[ 1.] [ 2.]] [[ 3.] [ 4.]] [[ 5.] [ 6.]]] x.shape = (3, 2, 1)

y = split(x, axis=1, num_outputs=2) // a list of 2 arrays with shape (3, 1, 1) y = [[[ 1.]] [[ 3.]] [[ 5.]]]

   [[[ 2.]]
    [[ 4.]]
    [[ 6.]]]

y[0].shape = (3, 1, 1)

z = split(x, axis=0, num_outputs=3) // a list of 3 arrays with shape (1, 2, 1) z = [[[ 1.] [ 2.]]]

   [[[ 3.]
     [ 4.]]]

   [[[ 5.]
     [ 6.]]]

z[0].shape = (1, 2, 1)

squeeze_axis=1 removes the axis with length 1 from the shapes of the output arrays. Note that setting squeeze_axis to $1$ removes axis with length 1 only along the axis which it is split. Also squeeze_axis can be set to true only if $input.shape[axis] == num_outputs$.

Example::

z = split(x, axis=0, num_outputs=3, squeeze_axis=1) // a list of 3 arrays with shape (2, 1) z = [[ 1.] [ 2.]]

   [[ 3.]
    [ 4.]]

   [[ 5.]
    [ 6.]]

z[0].shape = (2 ,1 )

Defined in src/operator/slice_channel.cc:L107

Arguments

data::NDArray-or-SymbolicNode: The input
num_outputs::int, required: Number of splits. Note that this should evenly divide the length of the axis.
axis::int, optional, default='1': Axis along which to split.
squeeze_axis::boolean, optional, default=0: If true, Removes the axis with length 1 from the shapes of the output arrays. Note that setting squeeze_axis to $true$ removes axis with length 1 only along the axis which it is split. Also squeeze_axis can be set to $true$ only if $input.shape[axis] == num_outputs$.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# Base.sqrt — Method.

sqrt(data)

Returns element-wise square-root value of the input.

.. math:: \textrm{sqrt}(x) = \sqrt{x}

Example::

sqrt([4, 9, 16]) = [2, 3, 4]

The storage type of $sqrt$ output depends upon the input storage type:

sqrt(default) = default
sqrt(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L564

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# Base.squeeze — Method.

squeeze(data, axis, num_args)

Remove single-dimensional entries from the shape of an array. Same behavior of defining the output tensor shape as numpy.squeeze for the most of cases. See the following note for exception.

Examples::

data = [[[0], [1], [2]]] squeeze(data) = [0, 1, 2] squeeze(data, axis=0) = [[0], [1], [2]] squeeze(data, axis=2) = [[0, 1, 2]] squeeze(data, axis=(0, 2)) = [0, 1, 2]

.. Note:: The output of this operator will keep at least one dimension not removed. For example, squeeze([[[4]]]) = [4], while in numpy.squeeze, the output will become a scalar.

Arguments

data::NDArray-or-SymbolicNode[]: data to squeeze
axis::int, optional, default='0': The axis in the result array along which the input arrays are stacked.
num_args::int, required: Number of inputs to be stacked.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# Base.sum — Method.

sum(data, axis, keepdims, exclude)

Computes the sum of array elements over given axes.

.. Note::

sum and sum_axis are equivalent. For ndarray of csr storage type summation along axis 0 and axis 1 is supported. Setting keepdims or exclude to True will cause a fallback to dense operator.

Example::

data = [[[1,2],[2,3],[1,3]], [[1,4],[4,3],[5,2]], [[7,1],[7,2],[7,3]]]

sum(data, axis=1) [[ 4. 8.] [ 10. 9.] [ 21. 6.]]

sum(data, axis=[1,2]) [ 12. 19. 27.]

data = [[1,2,0], [3,0,1], [4,1,0]]

csr = cast_storage(data, 'csr')

sum(csr, axis=0) [ 8. 3. 1.]

sum(csr, axis=1) [ 3. 4. 5.]

Defined in src/operator/tensor/broadcast_reduce_op_value.cc:L85

Arguments

data::NDArray-or-SymbolicNode: The input
axis::Shape(tuple), optional, default=[]: The axis or axes along which to perform the reduction.

`` The default,axis=(), will compute over all elements into a scalar array with shape(1,)`.

If axis is int, a reduction is performed on a particular axis.

If axis is a tuple of ints, a reduction is performed on all the axes specified in the tuple.

If exclude is true, reduction will be performed on the axes that are NOT in axis instead.

Negative values means indexing from right to left. `` *keepdims::boolean, optional, default=0: If this is set toTrue, the reduced axes are left in the result as dimension with size one. *exclude::boolean, optional, default=0: Whether to perform reduction on axis that are NOT in axis instead. *name::Symbol: The name of theSymbolicNode. (e.g.:my_symbol), optional. *attrs::Dict{Symbol, AbstractString}: The attributes associated with thisSymbolicNode`.

source

# Base.transpose — Method.

transpose(data, axes)

Permutes the dimensions of an array.

Examples::

x = [[ 1, 2], [ 3, 4]]

transpose(x) = [[ 1., 3.], [ 2., 4.]]

x = [[[ 1., 2.], [ 3., 4.]],

   [[ 5.,  6.],
    [ 7.,  8.]]]

transpose(x) = [[[ 1., 5.], [ 3., 7.]],

              [[ 2.,  6.],
               [ 4.,  8.]]]

transpose(x, axes=(1,0,2)) = [[[ 1., 2.], [ 5., 6.]],

                            [[ 3.,  4.],
                             [ 7.,  8.]]]

Defined in src/operator/tensor/matrix_op.cc:L257

Arguments

data::NDArray-or-SymbolicNode: Source input
axes::Shape(tuple), optional, default=[]: Target axis order. By default the axes will be inverted.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# Base.trunc — Method.

trunc(data)

Return the element-wise truncated value of the input.

The truncated value of the scalar x is the nearest integer i which is closer to zero than x is. In short, the fractional part of the signed number x is discarded.

Example::

trunc([-2.1, -1.9, 1.5, 1.9, 2.1]) = [-2., -1., 1., 1., 2.]

The storage type of $trunc$ output depends upon the input storage type:

trunc(default) = default
trunc(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L502

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.Activation — Method.

Activation(data, act_type)

Applies an activation function element-wise to the input.

The following activation functions are supported:

relu: Rectified Linear Unit, :math:y = max(x, 0)
sigmoid: :math:y = \frac{1}{1 + exp(-x)}
tanh: Hyperbolic tangent, :math:y = \frac{exp(x) - exp(-x)}{exp(x) + exp(-x)}
softrelu: Soft ReLU, or SoftPlus, :math:y = log(1 + exp(x))

Defined in src/operator/nn/activation.cc:L92

Arguments

data::NDArray-or-SymbolicNode: Input array to activation function.
act_type::{'relu', 'sigmoid', 'softrelu', 'tanh'}, required: Activation function to be applied.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.BatchNorm — Method.

BatchNorm(data, gamma, beta, moving_mean, moving_var, eps, momentum, fix_gamma, use_global_stats, output_mean_var, axis, cudnn_off)

Batch normalization.

Normalizes a data batch by mean and variance, and applies a scale $gamma$ as well as offset $beta$.

Assume the input has more than one dimension and we normalize along axis 1. We first compute the mean and variance along this axis:

.. math::

data_mean[i] = mean(data[:,i,:,...]) \ data_var[i] = var(data[:,i,:,...])

Then compute the normalized output, which has the same shape as input, as following:

.. math::

out[:,i,:,...] = \frac{data[:,i,:,...] - data_mean[i]}{\sqrt{data_var[i]+\epsilon}} * gamma[i] + beta[i]

Both mean and var returns a scalar by treating the input as a vector.

Assume the input has size k on axis 1, then both $gamma$ and $beta$ have shape (k,). If $output_mean_var$ is set to be true, then outputs both $data_mean$ and $data_var$ as well, which are needed for the backward pass.

Besides the inputs and the outputs, this operator accepts two auxiliary states, $moving_mean$ and $moving_var$, which are k-length vectors. They are global statistics for the whole dataset, which are updated by::

moving_mean = moving_mean * momentum + data_mean * (1 - momentum) moving_var = moving_var * momentum + data_var * (1 - momentum)

If $use_global_stats$ is set to be true, then $moving_mean$ and $moving_var$ are used instead of $data_mean$ and $data_var$ to compute the output. It is often used during inference.

The parameter $axis$ specifies which axis of the input shape denotes the 'channel' (separately normalized groups). The default is 1. Specifying -1 sets the channel axis to be the last item in the input shape.

Both $gamma$ and $beta$ are learnable parameters. But if $fix_gamma$ is true, then set $gamma$ to 1 and its gradient to 0.

Defined in src/operator/nn/batch_norm.cc:L400

Arguments

data::NDArray-or-SymbolicNode: Input data to batch normalization
gamma::NDArray-or-SymbolicNode: gamma array
beta::NDArray-or-SymbolicNode: beta array
moving_mean::NDArray-or-SymbolicNode: running mean of input
moving_var::NDArray-or-SymbolicNode: running variance of input
eps::double, optional, default=0.001: Epsilon to prevent div 0. Must be no less than CUDNN_BN_MIN_EPSILON defined in cudnn.h when using cudnn (usually 1e-5)
momentum::float, optional, default=0.9: Momentum for moving average
fix_gamma::boolean, optional, default=1: Fix gamma while training
use_global_stats::boolean, optional, default=0: Whether use global moving statistics instead of local batch-norm. This will force change batch-norm into a scale shift operator.
output_mean_var::boolean, optional, default=0: Output All,normal mean and var
axis::int, optional, default='1': Specify which shape axis the channel is specified
cudnn_off::boolean, optional, default=0: Do not select CUDNN operator, if available
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.BatchNorm_v1 — Method.

BatchNorm_v1(data, gamma, beta, eps, momentum, fix_gamma, use_global_stats, output_mean_var)

Batch normalization.

Normalizes a data batch by mean and variance, and applies a scale $gamma$ as well as offset $beta$.

Assume the input has more than one dimension and we normalize along axis 1. We first compute the mean and variance along this axis:

.. math::

data_mean[i] = mean(data[:,i,:,...]) \ data_var[i] = var(data[:,i,:,...])

Then compute the normalized output, which has the same shape as input, as following:

.. math::

out[:,i,:,...] = \frac{data[:,i,:,...] - data_mean[i]}{\sqrt{data_var[i]+\epsilon}} * gamma[i] + beta[i]

Both mean and var returns a scalar by treating the input as a vector.

Assume the input has size k on axis 1, then both $gamma$ and $beta$ have shape (k,). If $output_mean_var$ is set to be true, then outputs both $data_mean$ and $data_var$ as well, which are needed for the backward pass.

Besides the inputs and the outputs, this operator accepts two auxiliary states, $moving_mean$ and $moving_var$, which are k-length vectors. They are global statistics for the whole dataset, which are updated by::

moving_mean = moving_mean * momentum + data_mean * (1 - momentum) moving_var = moving_var * momentum + data_var * (1 - momentum)

If $use_global_stats$ is set to be true, then $moving_mean$ and $moving_var$ are used instead of $data_mean$ and $data_var$ to compute the output. It is often used during inference.

Both $gamma$ and $beta$ are learnable parameters. But if $fix_gamma$ is true, then set $gamma$ to 1 and its gradient to 0.

Defined in src/operator/batch_norm_v1.cc:L90

Arguments

data::NDArray-or-SymbolicNode: Input data to batch normalization
gamma::NDArray-or-SymbolicNode: gamma array
beta::NDArray-or-SymbolicNode: beta array
eps::float, optional, default=0.001: Epsilon to prevent div 0
momentum::float, optional, default=0.9: Momentum for moving average
fix_gamma::boolean, optional, default=1: Fix gamma while training
use_global_stats::boolean, optional, default=0: Whether use global moving statistics instead of local batch-norm. This will force change batch-norm into a scale shift operator.
output_mean_var::boolean, optional, default=0: Output All,normal mean and var
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.BilinearSampler — Method.

BilinearSampler(data, grid)

Applies bilinear sampling to input feature map.

Bilinear Sampling is the key of [NIPS2015] \"Spatial Transformer Networks\". The usage of the operator is very similar to remap function in OpenCV, except that the operator has the backward pass.

Given :math:data and :math:grid, then the output is computed by

.. math:: x_{src} = grid[batch, 0, y_{dst}, x_{dst}] \ y_{src} = grid[batch, 1, y_{dst}, x_{dst}] \ output[batch, channel, y_{dst}, x_{dst}] = G(data[batch, channel, y_{src}, x_{src})

:math:x_{dst}, :math:y_{dst} enumerate all spatial locations in :math:output, and :math:G() denotes the bilinear interpolation kernel. The out-boundary points will be padded with zeros.The shape of the output will be (data.shape[0], data.shape[1], grid.shape[2], grid.shape[3]).

The operator assumes that :math:data has 'NCHW' layout and :math:grid has been normalized to [-1, 1].

BilinearSampler often cooperates with GridGenerator which generates sampling grids for BilinearSampler. GridGenerator supports two kinds of transformation: $affine$ and $warp$. If users want to design a CustomOp to manipulate :math:grid, please firstly refer to the code of GridGenerator.

Example 1::

Zoom out data two times

data = array([[[[1, 4, 3, 6], [1, 8, 8, 9], [0, 4, 1, 5], [1, 0, 1, 3]]]])

affine_matrix = array([[2, 0, 0], [0, 2, 0]])

affine_matrix = reshape(affine_matrix, shape=(1, 6))

grid = GridGenerator(data=affine_matrix, transform_type='affine', target_shape=(4, 4))

out = BilinearSampler(data, grid)

out [[[[ 0, 0, 0, 0], [ 0, 3.5, 6.5, 0], [ 0, 1.25, 2.5, 0], [ 0, 0, 0, 0]]]

Example 2::

shift data horizontally by -1 pixel

data = array([[[[1, 4, 3, 6], [1, 8, 8, 9], [0, 4, 1, 5], [1, 0, 1, 3]]]])

warp_maxtrix = array([[[[1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1]], [[0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0]]]])

grid = GridGenerator(data=warp_matrix, transform_type='warp') out = BilinearSampler(data, grid)

out [[[[ 4, 3, 6, 0], [ 8, 8, 9, 0], [ 4, 1, 5, 0], [ 0, 1, 3, 0]]]

Defined in src/operator/bilinear_sampler.cc:L245

Arguments

data::NDArray-or-SymbolicNode: Input data to the BilinearsamplerOp.
grid::NDArray-or-SymbolicNode: Input grid to the BilinearsamplerOp.grid has two channels: x_src, y_src
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.BlockGrad — Method.

BlockGrad(data)

Stops gradient computation.

Stops the accumulated gradient of the inputs from flowing through this operator in the backward direction. In other words, this operator prevents the contribution of its inputs to be taken into account for computing gradients.

Example::

v1 = [1, 2] v2 = [0, 1] a = Variable('a') b = Variable('b') b_stop_grad = stop_gradient(3 * b) loss = MakeLoss(b_stop_grad + a)

executor = loss.simple_bind(ctx=cpu(), a=(1,2), b=(1,2)) executor.forward(is_train=True, a=v1, b=v2) executor.outputs [ 1. 5.]

executor.backward() executor.grad_arrays [ 0. 0.] [ 1. 1.]

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L166

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.Cast — Method.

Cast(data, dtype)

Casts all elements of the input to a new type.

.. note:: $Cast$ is deprecated. Use $cast$ instead.

Example::

cast([0.9, 1.3], dtype='int32') = [0, 1] cast([1e20, 11.1], dtype='float16') = [inf, 11.09375] cast([300, 11.1, 10.9, -1, -3], dtype='uint8') = [44, 11, 10, 255, 253]

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L310

Arguments

data::NDArray-or-SymbolicNode: The input.
dtype::{'float16', 'float32', 'float64', 'int32', 'uint8'}, required: Output data type.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.Concat — Method.

Concat(data, num_args, dim)

Note: Concat takes variable number of positional inputs. So instead of calling as Concat([x, y, z], num_args=3), one should call via Concat(x, y, z), and num_args will be determined automatically.

Joins input arrays along a given axis.

.. note:: Concat is deprecated. Use concat instead.

The dimensions of the input arrays should be the same except the axis along which they will be concatenated. The dimension of the output array along the concatenated axis will be equal to the sum of the corresponding dimensions of the input arrays.

Example::

x = [[1,1],[2,2]] y = [[3,3],[4,4],[5,5]] z = [[6,6], [7,7],[8,8]]

concat(x,y,z,dim=0) = [[ 1., 1.], [ 2., 2.], [ 3., 3.], [ 4., 4.], [ 5., 5.], [ 6., 6.], [ 7., 7.], [ 8., 8.]]

Note that you cannot concat x,y,z along dimension 1 since dimension 0 is not the same for all the input arrays.

concat(y,z,dim=1) = [[ 3., 3., 6., 6.], [ 4., 4., 7., 7.], [ 5., 5., 8., 8.]]

Defined in src/operator/concat.cc:L104

Arguments

data::NDArray-or-SymbolicNode[]: List of arrays to concatenate
num_args::int, required: Number of inputs to be concated.
dim::int, optional, default='1': the dimension to be concated.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.Convolution — Method.

Convolution(data, weight, bias, kernel, stride, dilate, pad, num_filter, num_group, workspace, no_bias, cudnn_tune, cudnn_off, layout)

Compute N-D convolution on (N+2)-D input.

In the 2-D convolution, given input data with shape (batch_size, channel, height, width), the output is computed by

.. math::

out[n,i,:,:] = bias[i] + \sum_{j=0}^{channel} data[n,j,:,:] \star weight[i,j,:,:]

where :math:\star is the 2-D cross-correlation operator.

For general 2-D convolution, the shapes are

data: (batch_size, channel, height, width)
weight: (num_filter, channel, kernel[0], kernel[1])
bias: (num_filter,)
out: (batch_size, num_filter, out_height, out_width).

Define::

f(x,k,p,s,d) = floor((x+2p-d(k-1)-1)/s)+1

then we have::

out_height=f(height, kernel[0], pad[0], stride[0], dilate[0]) out_width=f(width, kernel[1], pad[1], stride[1], dilate[1])

If $no_bias$ is set to be true, then the $bias$ term is ignored.

The default data $layout$ is NCHW, namely (batch_size, channel, height, width). We can choose other layouts such as NHWC.

If $num_group$ is larger than 1, denoted by g, then split the input $data$ evenly into g parts along the channel axis, and also evenly split $weight$ along the first dimension. Next compute the convolution on the i-th part of the data with the i-th weight part. The output is obtained by concatenating all the g results.

1-D convolution does not have height dimension but only width in space.

data: (batch_size, channel, width)
weight: (num_filter, channel, kernel[0])
bias: (num_filter,)
out: (batch_size, num_filter, out_width).

3-D convolution adds an additional depth dimension besides height and width. The shapes are

data: (batch_size, channel, depth, height, width)
weight: (num_filter, channel, kernel[0], kernel[1], kernel[2])
bias: (num_filter,)
out: (batch_size, num_filter, out_depth, out_height, out_width).

Both $weight$ and $bias$ are learnable parameters.

There are other options to tune the performance.

cudnn_tune: enable this option leads to higher startup time but may give faster speed. Options are
- off: no tuning
- limited_workspace:run test and pick the fastest algorithm that doesn't exceed workspace limit.
- fastest: pick the fastest algorithm and ignore workspace limit.
- None (default): the behavior is determined by environment variable $MXNET_CUDNN_AUTOTUNE_DEFAULT$. 0 for off, 1 for limited workspace (default), 2 for fastest.
- workspace: A large number leads to more (GPU) memory usage but may improve the performance.

Defined in src/operator/nn/convolution.cc:L170

Arguments

data::NDArray-or-SymbolicNode: Input data to the ConvolutionOp.
weight::NDArray-or-SymbolicNode: Weight matrix.
bias::NDArray-or-SymbolicNode: Bias parameter.
kernel::Shape(tuple), required: Convolution kernel size: (w,), (h, w) or (d, h, w)
stride::Shape(tuple), optional, default=[]: Convolution stride: (w,), (h, w) or (d, h, w). Defaults to 1 for each dimension.
dilate::Shape(tuple), optional, default=[]: Convolution dilate: (w,), (h, w) or (d, h, w). Defaults to 1 for each dimension.
pad::Shape(tuple), optional, default=[]: Zero pad for convolution: (w,), (h, w) or (d, h, w). Defaults to no padding.
num_filter::int (non-negative), required: Convolution filter(channel) number
num_group::int (non-negative), optional, default=1: Number of group partitions.
workspace::long (non-negative), optional, default=1024: Maximum temporary workspace allowed for convolution (MB).
no_bias::boolean, optional, default=0: Whether to disable bias parameter.
cudnn_tune::{None, 'fastest', 'limited_workspace', 'off'},optional, default='None': Whether to pick convolution algo by running performance test.
cudnn_off::boolean, optional, default=0: Turn off cudnn for this layer.
layout::{None, 'NCDHW', 'NCHW', 'NCW', 'NDHWC', 'NHWC'},optional, default='None': Set layout for input, output and weight. Empty for default layout: NCW for 1d, NCHW for 2d and NCDHW for 3d.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.Convolution_v1 — Method.

Convolution_v1(data, weight, bias, kernel, stride, dilate, pad, num_filter, num_group, workspace, no_bias, cudnn_tune, cudnn_off, layout)

This operator is DEPRECATED. Apply convolution to input then add a bias.

Arguments

data::NDArray-or-SymbolicNode: Input data to the ConvolutionV1Op.
weight::NDArray-or-SymbolicNode: Weight matrix.
bias::NDArray-or-SymbolicNode: Bias parameter.
kernel::Shape(tuple), required: convolution kernel size: (h, w) or (d, h, w)
stride::Shape(tuple), optional, default=[]: convolution stride: (h, w) or (d, h, w)
dilate::Shape(tuple), optional, default=[]: convolution dilate: (h, w) or (d, h, w)
pad::Shape(tuple), optional, default=[]: pad for convolution: (h, w) or (d, h, w)
num_filter::int (non-negative), required: convolution filter(channel) number
num_group::int (non-negative), optional, default=1: Number of group partitions. Equivalent to slicing input into num_group partitions, apply convolution on each, then concatenate the results
workspace::long (non-negative), optional, default=1024: Maximum tmp workspace allowed for convolution (MB).
no_bias::boolean, optional, default=0: Whether to disable bias parameter.
cudnn_tune::{None, 'fastest', 'limited_workspace', 'off'},optional, default='None': Whether to pick convolution algo by running performance test. Leads to higher startup time but may give faster speed. Options are: 'off': no tuning 'limited_workspace': run test and pick the fastest algorithm that doesn't exceed workspace limit. 'fastest': pick the fastest algorithm and ignore workspace limit. If set to None (default), behavior is determined by environment variable MXNET_CUDNN_AUTOTUNE_DEFAULT: 0 for off, 1 for limited workspace (default), 2 for fastest.
cudnn_off::boolean, optional, default=0: Turn off cudnn for this layer.
layout::{None, 'NCDHW', 'NCHW', 'NDHWC', 'NHWC'},optional, default='None': Set layout for input, output and weight. Empty for default layout: NCHW for 2d and NCDHW for 3d.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.Correlation — Method.

Correlation(data1, data2, kernel_size, max_displacement, stride1, stride2, pad_size, is_multiply)

Applies correlation to inputs.

The correlation layer performs multiplicative patch comparisons between two feature maps.

Given two multi-channel feature maps :math:f_{1}, f_{2}, with :math:w, :math:h, and :math:c being their width, height, and number of channels, the correlation layer lets the network compare each patch from :math:f_{1} with each patch from :math:f_{2}.

For now we consider only a single comparison of two patches. The 'correlation' of two patches centered at :math:x_{1} in the first map and :math:x_{2} in the second map is then defined as:

.. math:: c(x_{1}, x_{2}) = \sum_{o \in [-k,k] \times [-k,k]}

for a square patch of size :math:K:=2k+1.

Note that the equation above is identical to one step of a convolution in neural networks, but instead of convolving data with a filter, it convolves data with other data. For this reason, it has no training weights.

Computing :math:c(x_{1}, x_{2}) involves :math:c * K^{2} multiplications. Comparing all patch combinations involves :math:w^{2}*h^{2} such computations.

Given a maximum displacement :math:d, for each location :math:x_{1} it computes correlations :math:c(x_{1}, x_{2}) only in a neighborhood of size :math:D:=2d+1, by limiting the range of :math:x_{2}. We use strides :math:s_{1}, s_{2}, to quantize :math:x_{1} globally and to quantize :math:x_{2} within the neighborhood centered around :math:x_{1}.

The final output is defined by the following expression:

.. math:: out[n, q, i, j] = c(x_{i, j}, x_{q})

where :math:i and :math:j enumerate spatial locations in :math:f_{1}, and :math:q denotes the :math:q^{th} neighborhood of :math:x_{i,j}.

Defined in src/operator/correlation.cc:L192

Arguments

data1::NDArray-or-SymbolicNode: Input data1 to the correlation.
data2::NDArray-or-SymbolicNode: Input data2 to the correlation.
kernel_size::int (non-negative), optional, default=1: kernel size for Correlation must be an odd number
max_displacement::int (non-negative), optional, default=1: Max displacement of Correlation
stride1::int (non-negative), optional, default=1: stride1 quantize data1 globally
stride2::int (non-negative), optional, default=1: stride2 quantize data2 within the neighborhood centered around data1
pad_size::int (non-negative), optional, default=0: pad for Correlation
is_multiply::boolean, optional, default=1: operation type is either multiplication or subduction
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.Crop — Method.

Crop(data, num_args, offset, h_w, center_crop)

Note: Crop takes variable number of positional inputs. So instead of calling as Crop([x, y, z], num_args=3), one should call via Crop(x, y, z), and num_args will be determined automatically.

.. note:: Crop is deprecated. Use slice instead.

Crop the 2nd and 3rd dim of input data, with the corresponding size of h_w or with width and height of the second input symbol, i.e., with one input, we need h_w to specify the crop height and width, otherwise the second input symbol's size will be used

Defined in src/operator/crop.cc:L50

Arguments

data::SymbolicNode or SymbolicNode[]: Tensor or List of Tensors, the second input will be used as crop_like shape reference
num_args::int, required: Number of inputs for crop, if equals one, then we will use the h_wfor crop height and width, else if equals two, then we will use the heightand width of the second input symbol, we name crop_like here
offset::Shape(tuple), optional, default=[0,0]: crop offset coordinate: (y, x)
h_w::Shape(tuple), optional, default=[0,0]: crop height and width: (h, w)
center_crop::boolean, optional, default=0: If set to true, then it will use be the center_crop,or it will crop using the shape of crop_like
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.Custom — Method.

Custom(data, op_type)

Apply a custom operator implemented in a frontend language (like Python).

Custom operators should override required methods like forward and backward. The custom operator must be registered before it can be used. Please check the tutorial here: http://mxnet.io/faq/new_op.html.

Defined in src/operator/custom/custom.cc:L369

Arguments

data::NDArray-or-SymbolicNode[]: Input data for the custom operator.
op_type::string: Name of the custom operator. This is the name that is passed to mx.operator.register to register the operator.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.Deconvolution — Method.

Deconvolution(data, weight, bias, kernel, stride, dilate, pad, adj, target_shape, num_filter, num_group, workspace, no_bias, cudnn_tune, cudnn_off, layout)

Computes 1D or 2D transposed convolution (aka fractionally strided convolution) of the input tensor. This operation can be seen as the gradient of Convolution operation with respect to its input. Convolution usually reduces the size of the input. Transposed convolution works the other way, going from a smaller input to a larger output while preserving the connectivity pattern.

Arguments

data::NDArray-or-SymbolicNode: Input tensor to the deconvolution operation.
weight::NDArray-or-SymbolicNode: Weights representing the kernel.
bias::NDArray-or-SymbolicNode: Bias added to the result after the deconvolution operation.
kernel::Shape(tuple), required: Deconvolution kernel size: (w,), (h, w) or (d, h, w). This is same as the kernel size used for the corresponding convolution
stride::Shape(tuple), optional, default=[]: The stride used for the corresponding convolution: (w,), (h, w) or (d, h, w). Defaults to 1 for each dimension.
dilate::Shape(tuple), optional, default=[]: Dilation factor for each dimension of the input: (w,), (h, w) or (d, h, w). Defaults to 1 for each dimension.
pad::Shape(tuple), optional, default=[]: The amount of implicit zero padding added during convolution for each dimension of the input: (w,), (h, w) or (d, h, w). $(kernel-1)/2$ is usually a good choice. If target_shape is set, pad will be ignored and a padding that will generate the target shape will be used. Defaults to no padding.
adj::Shape(tuple), optional, default=[]: Adjustment for output shape: (w,), (h, w) or (d, h, w). If target_shape is set, adj will be ignored and computed accordingly.
target_shape::Shape(tuple), optional, default=[]: Shape of the output tensor: (w,), (h, w) or (d, h, w).
num_filter::int (non-negative), required: Number of output filters.
num_group::int (non-negative), optional, default=1: Number of groups partition.
workspace::long (non-negative), optional, default=512: Maximum temporal workspace allowed for deconvolution (MB).
no_bias::boolean, optional, default=1: Whether to disable bias parameter.
cudnn_tune::{None, 'fastest', 'limited_workspace', 'off'},optional, default='None': Whether to pick convolution algorithm by running performance test.
cudnn_off::boolean, optional, default=0: Turn off cudnn for this layer.
layout::{None, 'NCDHW', 'NCHW', 'NCW', 'NDHWC', 'NHWC'},optional, default='None': Set layout for input, output and weight. Empty for default layout, NCW for 1d, NCHW for 2d and NCDHW for 3d.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.Dropout — Method.

Dropout(data, p, mode)

Applies dropout operation to input array.

During training, each element of the input is set to zero with probability p. The whole array is rescaled by :math:1/(1-p) to keep the expected sum of the input unchanged.
During testing, this operator does not change the input if mode is 'training'. If mode is 'always', the same computaion as during training will be applied.

Example::

random.seed(998) input_array = array([[3., 0.5, -0.5, 2., 7.], [2., -0.4, 7., 3., 0.2]]) a = symbol.Variable('a') dropout = symbol.Dropout(a, p = 0.2) executor = dropout.simple_bind(a = input_array.shape)

If training

executor.forward(is_train = True, a = input_array) executor.outputs [[ 3.75 0.625 -0. 2.5 8.75 ] [ 2.5 -0.5 8.75 3.75 0. ]]

If testing

executor.forward(is_train = False, a = input_array) executor.outputs [[ 3. 0.5 -0.5 2. 7. ] [ 2. -0.4 7. 3. 0.2 ]]

Defined in src/operator/nn/dropout.cc:L79

Arguments

data::NDArray-or-SymbolicNode: Input array to which dropout will be applied.
p::float, optional, default=0.5: Fraction of the input that gets dropped out during training time.
mode::{'always', 'training'},optional, default='training': Whether to only turn on dropout during training or to also turn on for inference.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.ElementWiseSum — Method.

ElementWiseSum(args)

ElementWiseSum is an alias of add_n.

Note: ElementWiseSum takes variable number of positional inputs. So instead of calling as ElementWiseSum([x, y, z], num_args=3), one should call via ElementWiseSum(x, y, z), and num_args will be determined automatically.

Adds all input arguments element-wise.

.. math:: add_n(a_1, a_2, ..., a_n) = a_1 + a_2 + ... + a_n

$add_n$ is potentially more efficient than calling $add$ by n times.

The storage type of $add_n$ output depends on storage types of inputs

add_n(row_sparse, row_sparse, ..) = row_sparse
otherwise, $add_n$ generates output with default storage

Defined in src/operator/tensor/elemwise_sum.cc:L123

Arguments

args::NDArray-or-SymbolicNode[]: Positional input arguments
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.Embedding — Method.

Embedding(data, weight, input_dim, output_dim, dtype)

Maps integer indices to vector representations (embeddings).

This operator maps words to real-valued vectors in a high-dimensional space, called word embeddings. These embeddings can capture semantic and syntactic properties of the words. For example, it has been noted that in the learned embedding spaces, similar words tend to be close to each other and dissimilar words far apart.

For an input array of shape (d1, ..., dK), the shape of an output array is (d1, ..., dK, output_dim). All the input values should be integers in the range [0, input_dim).

If the input_dim is ip0 and output_dim is op0, then shape of the embedding weight matrix must be (ip0, op0).

By default, if any index mentioned is too large, it is replaced by the index that addresses the last vector in an embedding matrix.

Examples::

input_dim = 4 output_dim = 5

// Each row in weight matrix y represents a word. So, y = (w0,w1,w2,w3) y = [[ 0., 1., 2., 3., 4.], [ 5., 6., 7., 8., 9.], [ 10., 11., 12., 13., 14.], [ 15., 16., 17., 18., 19.]]

// Input array x represents n-grams(2-gram). So, x = [(w1,w3), (w0,w2)] x = [[ 1., 3.], [ 0., 2.]]

// Mapped input x to its vector representation y. Embedding(x, y, 4, 5) = [[[ 5., 6., 7., 8., 9.], [ 15., 16., 17., 18., 19.]],

                       [[  0.,   1.,   2.,   3.,   4.],
                        [ 10.,  11.,  12.,  13.,  14.]]]

Defined in src/operator/tensor/indexing_op.cc:L225

Arguments

data::NDArray-or-SymbolicNode: The input array to the embedding operator.
weight::NDArray-or-SymbolicNode: The embedding weight matrix.
input_dim::int, required: Vocabulary size of the input indices.
output_dim::int, required: Dimension of the embedding vectors.
dtype::{'float16', 'float32', 'float64', 'int32', 'uint8'},optional, default='float32': Data type of weight.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.FullyConnected — Method.

FullyConnected(data, weight, bias, num_hidden, no_bias, flatten)

Applies a linear transformation: :math:Y = XW^T + b.

If $flatten$ is set to be true, then the shapes are:

data: (batch_size, x1, x2, ..., xn)
weight: (num_hidden, x1 * x2 * ... * xn)
bias: (num_hidden,)
out: (batch_size, num_hidden)

If $flatten$ is set to be false, then the shapes are:

data: (x1, x2, ..., xn, input_dim)
weight: (num_hidden, input_dim)
bias: (num_hidden,)
out: (x1, x2, ..., xn, num_hidden)

The learnable parameters include both $weight$ and $bias$.

If $no_bias$ is set to be true, then the $bias$ term is ignored.

Defined in src/operator/nn/fully_connected.cc:L98

Arguments

data::NDArray-or-SymbolicNode: Input data.
weight::NDArray-or-SymbolicNode: Weight matrix.
bias::NDArray-or-SymbolicNode: Bias parameter.
num_hidden::int, required: Number of hidden nodes of the output.
no_bias::boolean, optional, default=0: Whether to disable bias parameter.
flatten::boolean, optional, default=1: Whether to collapse all but the first axis of the input data tensor.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.GridGenerator — Method.

GridGenerator(data, transform_type, target_shape)

Generates 2D sampling grid for bilinear sampling.

Arguments

data::NDArray-or-SymbolicNode: Input data to the function.
transform_type::{'affine', 'warp'}, required: The type of transformation. For affine, input data should be an affine matrix of size (batch, 6). For warp, input data should be an optical flow of size (batch, 2, h, w).
target_shape::Shape(tuple), optional, default=[0,0]: Specifies the output shape (H, W). This is required if transformation type is affine. If transformation type is warp, this parameter is ignored.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.Group — Method.

Group(nodes :: SymbolicNode...)

Create a SymbolicNode by grouping nodes together.

source

# MXNet.mx.IdentityAttachKLSparseReg — Method.

IdentityAttachKLSparseReg(data, sparseness_target, penalty, momentum)

Apply a sparse regularization to the output a sigmoid activation function.

Arguments

data::NDArray-or-SymbolicNode: Input data.
sparseness_target::float, optional, default=0.1: The sparseness target
penalty::float, optional, default=0.001: The tradeoff parameter for the sparseness penalty
momentum::float, optional, default=0.9: The momentum for running average
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.InstanceNorm — Method.

InstanceNorm(data, gamma, beta, eps)

Applies instance normalization to the n-dimensional input array.

This operator takes an n-dimensional input array where (n>2) and normalizes the input using the following formula:

.. math::

out = \frac{x - mean[data]}{ \sqrt{Var[data]} + \epsilon} * gamma + beta

This layer is similar to batch normalization layer (BatchNorm) with two differences: first, the normalization is carried out per example (instance), not over a batch. Second, the same normalization is applied both at test and train time. This operation is also known as contrast normalization.

If the input data is of shape [batch, channel, spacial_dim1, spacial_dim2, ...], gamma and beta parameters must be vectors of shape [channel].

This implementation is based on paper:

.. [1] Instance Normalization: The Missing Ingredient for Fast Stylization, D. Ulyanov, A. Vedaldi, V. Lempitsky, 2016 (arXiv:1607.08022v2).

Examples::

// Input of shape (2,1,2) x = [[[ 1.1, 2.2]], [[ 3.3, 4.4]]]

// gamma parameter of length 1 gamma = [1.5]

// beta parameter of length 1 beta = [0.5]

// Instance normalization is calculated with the above formula InstanceNorm(x,gamma,beta) = [[[-0.997527 , 1.99752665]], [[-0.99752653, 1.99752724]]]

Defined in src/operator/instance_norm.cc:L95

Arguments

data::NDArray-or-SymbolicNode: An n-dimensional input array (n > 2) of the form [batch, channel, spatial_dim1, spatial_dim2, ...].
gamma::NDArray-or-SymbolicNode: A vector of length 'channel', which multiplies the normalized input.
beta::NDArray-or-SymbolicNode: A vector of length 'channel', which is added to the product of the normalized input and the weight.
eps::float, optional, default=0.001: An epsilon parameter to prevent division by 0.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.L2Normalization — Method.

L2Normalization(data, eps, mode)

Normalize the input array using the L2 norm.

For 1-D NDArray, it computes::

out = data / sqrt(sum(data ** 2) + eps)

For N-D NDArray, if the input array has shape (N, N, ..., N),

with $mode$ = $instance$, it normalizes each instance in the multidimensional array by its L2 norm.::

for i in 0...N out[i,:,:,...,:] = data[i,:,:,...,:] / sqrt(sum(data[i,:,:,...,:] ** 2) + eps)

with $mode$ = $channel$, it normalizes each channel in the array by its L2 norm.::

for i in 0...N out[:,i,:,...,:] = data[:,i,:,...,:] / sqrt(sum(data[:,i,:,...,:] ** 2) + eps)

with $mode$ = $spatial$, it normalizes the cross channel norm for each position in the array by its L2 norm.::

for dim in 2...N for i in 0...N out[.....,i,...] = take(out, indices=i, axis=dim) / sqrt(sum(take(out, indices=i, axis=dim) ** 2) + eps) -dim-

Example::

x = [[[1,2], [3,4]], [[2,2], [5,6]]]

L2Normalization(x, mode='instance') =[[[ 0.18257418 0.36514837] [ 0.54772252 0.73029673]] [[ 0.24077171 0.24077171] [ 0.60192931 0.72231513]]]

L2Normalization(x, mode='channel') =[[[ 0.31622776 0.44721359] [ 0.94868326 0.89442718]] [[ 0.37139067 0.31622776] [ 0.92847669 0.94868326]]]

L2Normalization(x, mode='spatial') =[[[ 0.44721359 0.89442718] [ 0.60000002 0.80000001]] [[ 0.70710677 0.70710677] [ 0.6401844 0.76822126]]]

Defined in src/operator/l2_normalization.cc:L93

Arguments

data::NDArray-or-SymbolicNode: Input array to normalize.
eps::float, optional, default=1e-10: A small constant for numerical stability.
mode::{'channel', 'instance', 'spatial'},optional, default='instance': Specify the dimension along which to compute L2 norm.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.LRN — Method.

LRN(data, alpha, beta, knorm, nsize)

Applies local response normalization to the input.

The local response normalization layer performs "lateral inhibition" by normalizing over local input regions.

If :math:a_{x,y}^{i} is the activity of a neuron computed by applying kernel :math:i at position :math:(x, y) and then applying the ReLU nonlinearity, the response-normalized activity :math:b_{x,y}^{i} is given by the expression:

.. math:: b_{x,y}^{i} = \frac{a_{x,y}^{i}}{\Bigg({k + \alpha \sum_{j=max(0, i-\frac{n}{2})}^{min(N-1, i+\frac{n}{2})} (a_{x,y}^{j})^{2}}\Bigg)^{\beta}}

where the sum runs over :math:n "adjacent" kernel maps at the same spatial position, and :math:N is the total number of kernels in the layer.

Defined in src/operator/lrn.cc:L73

Arguments

data::NDArray-or-SymbolicNode: Input data.
alpha::float, optional, default=0.0001: The variance scaling parameter :math:lpha in the LRN expression.
beta::float, optional, default=0.75: The power parameter :math:eta in the LRN expression.
knorm::float, optional, default=2: The parameter :math:k in the LRN expression.
nsize::int (non-negative), required: normalization window width in elements.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.LeakyReLU — Method.

LeakyReLU(data, gamma, act_type, slope, lower_bound, upper_bound)

Applies Leaky rectified linear unit activation element-wise to the input.

Leaky ReLUs attempt to fix the "dying ReLU" problem by allowing a small slope when the input is negative and has a slope of one when input is positive.

The following modified ReLU Activation functions are supported:

elu: Exponential Linear Unit. y = x > 0 ? x : slope * (exp(x)-1)
leaky: Leaky ReLU. y = x > 0 ? x : slope * x
prelu: Parametric ReLU. This is same as leaky except that slope is learnt during training.
rrelu: Randomized ReLU. same as leaky but the slope is uniformly and randomly chosen from [lower_bound, upper_bound) for training, while fixed to be (lower_bound+upper_bound)/2 for inference.

Defined in src/operator/leaky_relu.cc:L58

Arguments

data::NDArray-or-SymbolicNode: Input data to activation function.
gamma::NDArray-or-SymbolicNode: Slope parameter for PReLU. Only required when act_type is 'prelu'. It should be either a vector of size 1, or the same size as the second dimension of data.
act_type::{'elu', 'leaky', 'prelu', 'rrelu'},optional, default='leaky': Activation function to be applied.
slope::float, optional, default=0.25: Init slope for the activation. (For leaky and elu only)
lower_bound::float, optional, default=0.125: Lower bound of random slope. (For rrelu only)
upper_bound::float, optional, default=0.334: Upper bound of random slope. (For rrelu only)
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.LinearRegressionOutput — Method.

LinearRegressionOutput(data, label, grad_scale)

Computes and optimizes for squared loss during backward propagation. Just outputs $data$ during forward propagation.

If :math:\hat{y}_i is the predicted value of the i-th sample, and :math:y_i is the corresponding target value, then the squared loss estimated over :math:n samples is defined as

:math:\text{SquaredLoss}(\textbf{Y}, \hat{\textbf{Y}} ) = \frac{1}{n} \sum_{i=0}^{n-1} \lVert \textbf{y}_i - \hat{\textbf{y}}_i \rVert_2

.. note:: Use the LinearRegressionOutput as the final output layer of a net.

By default, gradients of this loss function are scaled by factor 1/m, where m is the number of regression outputs of a training example. The parameter grad_scale can be used to change this scale to grad_scale/m.

Defined in src/operator/regression_output.cc:L80

Arguments

data::NDArray-or-SymbolicNode: Input data to the function.
label::NDArray-or-SymbolicNode: Input label to the function.
grad_scale::float, optional, default=1: Scale the gradient by a float factor
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.LogisticRegressionOutput — Method.

LogisticRegressionOutput(data, label, grad_scale)

Applies a logistic function to the input.

The logistic function, also known as the sigmoid function, is computed as :math:\frac{1}{1+exp(-\textbf{x})}.

Commonly, the sigmoid is used to squash the real-valued output of a linear model :math:wTx+b into the [0,1] range so that it can be interpreted as a probability. It is suitable for binary classification or probability prediction tasks.

.. note:: Use the LogisticRegressionOutput as the final output layer of a net.

By default, gradients of this loss function are scaled by factor 1/m, where m is the number of regression outputs of a training example. The parameter grad_scale can be used to change this scale to grad_scale/m.

Defined in src/operator/regression_output.cc:L122

Arguments

data::NDArray-or-SymbolicNode: Input data to the function.
label::NDArray-or-SymbolicNode: Input label to the function.
grad_scale::float, optional, default=1: Scale the gradient by a float factor
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.MAERegressionOutput — Method.

MAERegressionOutput(data, label, grad_scale)

Computes mean absolute error of the input.

MAE is a risk metric corresponding to the expected value of the absolute error.

If :math:\hat{y}_i is the predicted value of the i-th sample, and :math:y_i is the corresponding target value, then the mean absolute error (MAE) estimated over :math:n samples is defined as

:math:\text{MAE}(\textbf{Y}, \hat{\textbf{Y}} ) = \frac{1}{n} \sum_{i=0}^{n-1} \lVert \textbf{y}_i - \hat{\textbf{y}}_i \rVert_1

.. note:: Use the MAERegressionOutput as the final output layer of a net.

By default, gradients of this loss function are scaled by factor 1/m, where m is the number of regression outputs of a training example. The parameter grad_scale can be used to change this scale to grad_scale/m.

Defined in src/operator/regression_output.cc:L101

Arguments

data::NDArray-or-SymbolicNode: Input data to the function.
label::NDArray-or-SymbolicNode: Input label to the function.
grad_scale::float, optional, default=1: Scale the gradient by a float factor
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.MakeLoss — Method.

MakeLoss(data, grad_scale, valid_thresh, normalization)

Make your own loss function in network construction.

This operator accepts a customized loss function symbol as a terminal loss and the symbol should be an operator with no backward dependency. The output of this function is the gradient of loss with respect to the input data.

For example, if you are a making a cross entropy loss function. Assume $out$ is the predicted output and $label$ is the true label, then the cross entropy can be defined as::

cross_entropy = label * log(out) + (1 - label) * log(1 - out) loss = MakeLoss(cross_entropy)

We will need to use $MakeLoss$ when we are creating our own loss function or we want to combine multiple loss functions. Also we may want to stop some variables' gradients from backpropagation. See more detail in $BlockGrad$ or $stop_gradient$.

In addition, we can give a scale to the loss by setting $grad_scale$, so that the gradient of the loss will be rescaled in the backpropagation.

.. note:: This operator should be used as a Symbol instead of NDArray.

Defined in src/operator/make_loss.cc:L71

Arguments

data::NDArray-or-SymbolicNode: Input array.
grad_scale::float, optional, default=1: Gradient scale as a supplement to unary and binary operators
valid_thresh::float, optional, default=0: clip each element in the array to 0 when it is less than $valid_thresh$. This is used when $normalization$ is set to $'valid'$.
normalization::{'batch', 'null', 'valid'},optional, default='null': If this is set to null, the output gradient will not be normalized. If this is set to batch, the output gradient will be divided by the batch size. If this is set to valid, the output gradient will be divided by the number of valid input elements.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.Pad — Method.

Pad(data, mode, pad_width, constant_value)

Pads an input array with a constant or edge values of the array.

.. note:: Pad is deprecated. Use pad instead.

.. note:: Current implementation only supports 4D and 5D input arrays with padding applied only on axes 1, 2 and 3. Expects axes 4 and 5 in pad_width to be zero.

This operation pads an input array with either a constant_value or edge values along each axis of the input array. The amount of padding is specified by pad_width.

pad_width is a tuple of integer padding widths for each axis of the format $(before_1, after_1, ... , before_N, after_N)$. The pad_width should be of length $2*N$ where $N$ is the number of dimensions of the array.

For dimension $N$ of the input array, $before_N$ and $after_N$ indicates how many values to add before and after the elements of the array along dimension $N$. The widths of the higher two dimensions $before_1$, $after_1$, $before_2$, $after_2$ must be 0.

Example::

x = [[[[ 1. 2. 3.] [ 4. 5. 6.]]

     [[  7.   8.   9.]
      [ 10.  11.  12.]]]


    [[[ 11.  12.  13.]
      [ 14.  15.  16.]]

     [[ 17.  18.  19.]
      [ 20.  21.  22.]]]]

pad(x,mode="edge", pad_width=(0,0,0,0,1,1,1,1)) =

     [[[[  1.   1.   2.   3.   3.]
        [  1.   1.   2.   3.   3.]
        [  4.   4.   5.   6.   6.]
        [  4.   4.   5.   6.   6.]]

       [[  7.   7.   8.   9.   9.]
        [  7.   7.   8.   9.   9.]
        [ 10.  10.  11.  12.  12.]
        [ 10.  10.  11.  12.  12.]]]


      [[[ 11.  11.  12.  13.  13.]
        [ 11.  11.  12.  13.  13.]
        [ 14.  14.  15.  16.  16.]
        [ 14.  14.  15.  16.  16.]]

       [[ 17.  17.  18.  19.  19.]
        [ 17.  17.  18.  19.  19.]
        [ 20.  20.  21.  22.  22.]
        [ 20.  20.  21.  22.  22.]]]]

pad(x, mode="constant", constant_value=0, pad_width=(0,0,0,0,1,1,1,1)) =

     [[[[  0.   0.   0.   0.   0.]
        [  0.   1.   2.   3.   0.]
        [  0.   4.   5.   6.   0.]
        [  0.   0.   0.   0.   0.]]

       [[  0.   0.   0.   0.   0.]
        [  0.   7.   8.   9.   0.]
        [  0.  10.  11.  12.   0.]
        [  0.   0.   0.   0.   0.]]]


      [[[  0.   0.   0.   0.   0.]
        [  0.  11.  12.  13.   0.]
        [  0.  14.  15.  16.   0.]
        [  0.   0.   0.   0.   0.]]

       [[  0.   0.   0.   0.   0.]
        [  0.  17.  18.  19.   0.]
        [  0.  20.  21.  22.   0.]
        [  0.   0.   0.   0.   0.]]]]

Defined in src/operator/pad.cc:L766

Arguments

data::NDArray-or-SymbolicNode: An n-dimensional input array.
mode::{'constant', 'edge', 'reflect'}, required: Padding type to use. "constant" pads with constant_value "edge" pads using the edge values of the input array "reflect" pads by reflecting values with respect to the edges.
pad_width::Shape(tuple), required: Widths of the padding regions applied to the edges of each axis. It is a tuple of integer padding widths for each axis of the format $(before_1, after_1, ... , before_N, after_N)$. It should be of length $2*N$ where $N$ is the number of dimensions of the array.This is equivalent to pad_width in numpy.pad, but flattened.
constant_value::double, optional, default=0: The value used for padding when mode is "constant".
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.Pooling — Method.

Pooling(data, global_pool, cudnn_off, kernel, pool_type, pooling_convention, stride, pad)

Performs pooling on the input.

The shapes for 1-D pooling are

data: (batch_size, channel, width),
out: (batch_size, num_filter, out_width).

The shapes for 2-D pooling are

data: (batch_size, channel, height, width)
out: (batch_size, num_filter, out_height, out_width), with::

out_height = f(height, kernel[0], pad[0], stride[0]) out_width = f(width, kernel[1], pad[1], stride[1])

The definition of f depends on $pooling_convention$, which has two options:

valid (default)::

f(x, k, p, s) = floor((x+2*p-k)/s)+1 * full, which is compatible with Caffe::

f(x, k, p, s) = ceil((x+2*p-k)/s)+1

But $global_pool$ is set to be true, then do a global pooling, namely reset $kernel=(height, width)$.

Three pooling options are supported by $pool_type$:

avg: average pooling
max: max pooling
sum: sum pooling

For 3-D pooling, an additional depth dimension is added before height. Namely the input data will have shape (batch_size, channel, depth, height, width).

Defined in src/operator/nn/pooling.cc:L133

Arguments

data::NDArray-or-SymbolicNode: Input data to the pooling operator.
global_pool::boolean, optional, default=0: Ignore kernel size, do global pooling based on current input feature map.
cudnn_off::boolean, optional, default=0: Turn off cudnn pooling and use MXNet pooling operator.
kernel::Shape(tuple), required: Pooling kernel size: (y, x) or (d, y, x)
pool_type::{'avg', 'max', 'sum'}, required: Pooling type to be applied.
pooling_convention::{'full', 'valid'},optional, default='valid': Pooling convention to be applied.
stride::Shape(tuple), optional, default=[]: Stride: for pooling (y, x) or (d, y, x). Defaults to 1 for each dimension.
pad::Shape(tuple), optional, default=[]: Pad for pooling: (y, x) or (d, y, x). Defaults to no padding.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.Pooling_v1 — Method.

Pooling_v1(data, global_pool, kernel, pool_type, pooling_convention, stride, pad)

This operator is DEPRECATED. Perform pooling on the input.

The shapes for 2-D pooling is

data: (batch_size, channel, height, width)
out: (batch_size, num_filter, out_height, out_width), with::

out_height = f(height, kernel[0], pad[0], stride[0]) out_width = f(width, kernel[1], pad[1], stride[1])

The definition of f depends on $pooling_convention$, which has two options:

valid (default)::

f(x, k, p, s) = floor((x+2*p-k)/s)+1 * full, which is compatible with Caffe::

f(x, k, p, s) = ceil((x+2*p-k)/s)+1

But $global_pool$ is set to be true, then do a global pooling, namely reset $kernel=(height, width)$.

Three pooling options are supported by $pool_type$:

avg: average pooling
max: max pooling
sum: sum pooling

1-D pooling is special case of 2-D pooling with weight=1 and kernel[1]=1.

For 3-D pooling, an additional depth dimension is added before height. Namely the input data will have shape (batch_size, channel, depth, height, width).

Defined in src/operator/pooling_v1.cc:L104

Arguments

data::NDArray-or-SymbolicNode: Input data to the pooling operator.
global_pool::boolean, optional, default=0: Ignore kernel size, do global pooling based on current input feature map.
kernel::Shape(tuple), required: pooling kernel size: (y, x) or (d, y, x)
pool_type::{'avg', 'max', 'sum'}, required: Pooling type to be applied.
pooling_convention::{'full', 'valid'},optional, default='valid': Pooling convention to be applied.
stride::Shape(tuple), optional, default=[]: stride: for pooling (y, x) or (d, y, x)
pad::Shape(tuple), optional, default=[]: pad for pooling: (y, x) or (d, y, x)
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.RNN — Method.

RNN(data, parameters, state, state_cell, state_size, num_layers, bidirectional, mode, p, state_outputs)

Applies a recurrent layer to input.

Arguments

data::NDArray-or-SymbolicNode: Input data to RNN
parameters::NDArray-or-SymbolicNode: Vector of all RNN trainable parameters concatenated
state::NDArray-or-SymbolicNode: initial hidden state of the RNN
state_cell::NDArray-or-SymbolicNode: initial cell state for LSTM networks (only for LSTM)
state_size::int (non-negative), required: size of the state for each layer
num_layers::int (non-negative), required: number of stacked layers
bidirectional::boolean, optional, default=0: whether to use bidirectional recurrent layers
mode::{'gru', 'lstm', 'rnn_relu', 'rnn_tanh'}, required: the type of RNN to compute
p::float, optional, default=0: Dropout probability, fraction of the input that gets dropped out at training time
state_outputs::boolean, optional, default=0: Whether to have the states as symbol outputs.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.ROIPooling — Method.

ROIPooling(data, rois, pooled_size, spatial_scale)

Performs region of interest(ROI) pooling on the input array.

ROI pooling is a variant of a max pooling layer, in which the output size is fixed and region of interest is a parameter. Its purpose is to perform max pooling on the inputs of non-uniform sizes to obtain fixed-size feature maps. ROI pooling is a neural-net layer mostly used in training a Fast R-CNN network for object detection.

This operator takes a 4D feature map as an input array and region proposals as rois, then it pools over sub-regions of input and produces a fixed-sized output array regardless of the ROI size.

To crop the feature map accordingly, you can resize the bounding box coordinates by changing the parameters rois and spatial_scale.

The cropped feature maps are pooled by standard max pooling operation to a fixed size output indicated by a pooled_size parameter. batch_size will change to the number of region bounding boxes after ROIPooling.

The size of each region of interest doesn't have to be perfectly divisible by the number of pooling sections(pooled_size).

Example::

x = [[[[ 0., 1., 2., 3., 4., 5.], [ 6., 7., 8., 9., 10., 11.], [ 12., 13., 14., 15., 16., 17.], [ 18., 19., 20., 21., 22., 23.], [ 24., 25., 26., 27., 28., 29.], [ 30., 31., 32., 33., 34., 35.], [ 36., 37., 38., 39., 40., 41.], [ 42., 43., 44., 45., 46., 47.]]]]

// region of interest i.e. bounding box coordinates. y = [[0,0,0,4,4]]

// returns array of shape (2,2) according to the given roi with max pooling. ROIPooling(x, y, (2,2), 1.0) = [[[[ 14., 16.], [ 26., 28.]]]]

// region of interest is changed due to the change in spacial_scale parameter. ROIPooling(x, y, (2,2), 0.7) = [[[[ 7., 9.], [ 19., 21.]]]]

Defined in src/operator/roi_pooling.cc:L287

Arguments

data::NDArray-or-SymbolicNode: The input array to the pooling operator, a 4D Feature maps
rois::NDArray-or-SymbolicNode: Bounding box coordinates, a 2D array of [[batch_index, x1, y1, x2, y2]], where (x1, y1) and (x2, y2) are top left and bottom right corners of designated region of interest. batch_index indicates the index of corresponding image in the input array
pooled_size::Shape(tuple), required: ROI pooling output shape (h,w)
spatial_scale::float, required: Ratio of input feature map height (or w) to raw image height (or w). Equals the reciprocal of total stride in convolutional layers
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.SVMOutput — Method.

SVMOutput(data, label, margin, regularization_coefficient, use_linear)

Computes support vector machine based transformation of the input.

This tutorial demonstrates using SVM as output layer for classification instead of softmax: https://github.com/dmlc/mxnet/tree/master/example/svm_mnist.

Arguments

data::NDArray-or-SymbolicNode: Input data for SVM transformation.
label::NDArray-or-SymbolicNode: Class label for the input data.
margin::float, optional, default=1: The loss function penalizes outputs that lie outside this margin. Default margin is 1.
regularization_coefficient::float, optional, default=1: Regularization parameter for the SVM. This balances the tradeoff between coefficient size and error.
use_linear::boolean, optional, default=0: Whether to use L1-SVM objective. L2-SVM objective is used by default.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.SequenceLast — Method.

SequenceLast(data, sequence_length, use_sequence_length, axis)

Takes the last element of a sequence.

This function takes an n-dimensional input array of the form [max_sequence_length, batch_size, other_feature_dims] and returns a (n-1)-dimensional array of the form [batch_size, other_feature_dims].

Parameter sequence_length is used to handle variable-length sequences. sequence_length should be an input array of positive ints of dimension [batch_size]. To use this parameter, set use_sequence_length to True, otherwise each example in the batch is assumed to have the max sequence length.

.. note:: Alternatively, you can also use take operator.

Example::

x = [[[ 1., 2., 3.], [ 4., 5., 6.], [ 7., 8., 9.]],

    [[ 10.,   11.,   12.],
     [ 13.,   14.,   15.],
     [ 16.,   17.,   18.]],

    [[  19.,   20.,   21.],
     [  22.,   23.,   24.],
     [  25.,   26.,   27.]]]

// returns last sequence when sequence_length parameter is not used SequenceLast(x) = [[ 19., 20., 21.], [ 22., 23., 24.], [ 25., 26., 27.]]

// sequence_length is used SequenceLast(x, sequence_length=[1,1,1], use_sequence_length=True) = [[ 1., 2., 3.], [ 4., 5., 6.], [ 7., 8., 9.]]

// sequence_length is used SequenceLast(x, sequence_length=[1,2,3], use_sequence_length=True) = [[ 1., 2., 3.], [ 13., 14., 15.], [ 25., 26., 27.]]

Defined in src/operator/sequence_last.cc:L92

Arguments

data::NDArray-or-SymbolicNode: n-dimensional input array of the form [max_sequence_length, batch_size, other_feature_dims] where n>2
sequence_length::NDArray-or-SymbolicNode: vector of sequence lengths of the form [batch_size]
use_sequence_length::boolean, optional, default=0: If set to true, this layer takes in an extra input parameter sequence_length to specify variable length sequence
axis::int, optional, default='0': The sequence axis. Only values of 0 and 1 are currently supported.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.SequenceMask — Method.

SequenceMask(data, sequence_length, use_sequence_length, value, axis)

Sets all elements outside the sequence to a constant value.

This function takes an n-dimensional input array of the form [max_sequence_length, batch_size, other_feature_dims] and returns an array of the same shape.

Parameter sequence_length is used to handle variable-length sequences. sequence_length should be an input array of positive ints of dimension [batch_size]. To use this parameter, set use_sequence_length to True, otherwise each example in the batch is assumed to have the max sequence length and this operator works as the identity operator.

Example::

x = [[[ 1., 2., 3.], [ 4., 5., 6.]],

    [[  7.,   8.,   9.],
     [ 10.,  11.,  12.]],

    [[ 13.,  14.,   15.],
     [ 16.,  17.,   18.]]]

// Batch 1 B1 = [[ 1., 2., 3.], [ 7., 8., 9.], [ 13., 14., 15.]]

// Batch 2 B2 = [[ 4., 5., 6.], [ 10., 11., 12.], [ 16., 17., 18.]]

// works as identity operator when sequence_length parameter is not used SequenceMask(x) = [[[ 1., 2., 3.], [ 4., 5., 6.]],

                  [[  7.,   8.,   9.],
                   [ 10.,  11.,  12.]],

                  [[ 13.,  14.,   15.],
                   [ 16.,  17.,   18.]]]

// sequence_length [1,1] means 1 of each batch will be kept // and other rows are masked with default mask value = 0 SequenceMask(x, sequence_length=[1,1], use_sequence_length=True) = [[[ 1., 2., 3.], [ 4., 5., 6.]],

             [[  0.,   0.,   0.],
              [  0.,   0.,   0.]],

             [[  0.,   0.,   0.],
              [  0.,   0.,   0.]]]

// sequence_length [2,3] means 2 of batch B1 and 3 of batch B2 will be kept // and other rows are masked with value = 1 SequenceMask(x, sequence_length=[2,3], use_sequence_length=True, value=1) = [[[ 1., 2., 3.], [ 4., 5., 6.]],

             [[  7.,   8.,   9.],
              [  10.,  11.,  12.]],

             [[   1.,   1.,   1.],
              [  16.,  17.,  18.]]]

Defined in src/operator/sequence_mask.cc:L114

Arguments

data::NDArray-or-SymbolicNode: n-dimensional input array of the form [max_sequence_length, batch_size, other_feature_dims] where n>2
sequence_length::NDArray-or-SymbolicNode: vector of sequence lengths of the form [batch_size]
use_sequence_length::boolean, optional, default=0: If set to true, this layer takes in an extra input parameter sequence_length to specify variable length sequence
value::float, optional, default=0: The value to be used as a mask.
axis::int, optional, default='0': The sequence axis. Only values of 0 and 1 are currently supported.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.SequenceReverse — Method.

SequenceReverse(data, sequence_length, use_sequence_length, axis)

Reverses the elements of each sequence.

This function takes an n-dimensional input array of the form [max_sequence_length, batch_size, other_feature_dims] and returns an array of the same shape.

Parameter sequence_length is used to handle variable-length sequences. sequence_length should be an input array of positive ints of dimension [batch_size]. To use this parameter, set use_sequence_length to True, otherwise each example in the batch is assumed to have the max sequence length.

Example::

x = [[[ 1., 2., 3.], [ 4., 5., 6.]],

    [[  7.,   8.,   9.],
     [ 10.,  11.,  12.]],

    [[ 13.,  14.,   15.],
     [ 16.,  17.,   18.]]]

// Batch 1 B1 = [[ 1., 2., 3.], [ 7., 8., 9.], [ 13., 14., 15.]]

// Batch 2 B2 = [[ 4., 5., 6.], [ 10., 11., 12.], [ 16., 17., 18.]]

// returns reverse sequence when sequence_length parameter is not used SequenceReverse(x) = [[[ 13., 14., 15.], [ 16., 17., 18.]],

                     [[  7.,   8.,   9.],
                      [ 10.,  11.,  12.]],

                     [[  1.,   2.,   3.],
                      [  4.,   5.,   6.]]]

// sequence_length [2,2] means 2 rows of // both batch B1 and B2 will be reversed. SequenceReverse(x, sequence_length=[2,2], use_sequence_length=True) = [[[ 7., 8., 9.], [ 10., 11., 12.]],

                  [[  1.,   2.,   3.],
                   [  4.,   5.,   6.]],

                  [[ 13.,  14.,   15.],
                   [ 16.,  17.,   18.]]]

// sequence_length [2,3] means 2 of batch B2 and 3 of batch B3 // will be reversed. SequenceReverse(x, sequence_length=[2,3], use_sequence_length=True) = [[[ 7., 8., 9.], [ 16., 17., 18.]],

                 [[  1.,   2.,   3.],
                  [ 10.,  11.,  12.]],

                 [[ 13.,  14,   15.],
                  [  4.,   5.,   6.]]]

Defined in src/operator/sequence_reverse.cc:L113

Arguments

data::NDArray-or-SymbolicNode: n-dimensional input array of the form [max_sequence_length, batch_size, other dims] where n>2
sequence_length::NDArray-or-SymbolicNode: vector of sequence lengths of the form [batch_size]
use_sequence_length::boolean, optional, default=0: If set to true, this layer takes in an extra input parameter sequence_length to specify variable length sequence
axis::int, optional, default='0': The sequence axis. Only 0 is currently supported.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.SliceChannel — Method.

SliceChannel(data, num_outputs, axis, squeeze_axis)

Splits an array along a particular axis into multiple sub-arrays.

.. note:: $SliceChannel$ is deprecated. Use $split$ instead.

Note that num_outputs should evenly divide the length of the axis along which to split the array.

Example::

x = [[[ 1.] [ 2.]] [[ 3.] [ 4.]] [[ 5.] [ 6.]]] x.shape = (3, 2, 1)

y = split(x, axis=1, num_outputs=2) // a list of 2 arrays with shape (3, 1, 1) y = [[[ 1.]] [[ 3.]] [[ 5.]]]

   [[[ 2.]]
    [[ 4.]]
    [[ 6.]]]

y[0].shape = (3, 1, 1)

z = split(x, axis=0, num_outputs=3) // a list of 3 arrays with shape (1, 2, 1) z = [[[ 1.] [ 2.]]]

   [[[ 3.]
     [ 4.]]]

   [[[ 5.]
     [ 6.]]]

z[0].shape = (1, 2, 1)

squeeze_axis=1 removes the axis with length 1 from the shapes of the output arrays. Note that setting squeeze_axis to $1$ removes axis with length 1 only along the axis which it is split. Also squeeze_axis can be set to true only if $input.shape[axis] == num_outputs$.

Example::

z = split(x, axis=0, num_outputs=3, squeeze_axis=1) // a list of 3 arrays with shape (2, 1) z = [[ 1.] [ 2.]]

   [[ 3.]
    [ 4.]]

   [[ 5.]
    [ 6.]]

z[0].shape = (2 ,1 )

Defined in src/operator/slice_channel.cc:L107

Arguments

data::NDArray-or-SymbolicNode: The input
num_outputs::int, required: Number of splits. Note that this should evenly divide the length of the axis.
axis::int, optional, default='1': Axis along which to split.
squeeze_axis::boolean, optional, default=0: If true, Removes the axis with length 1 from the shapes of the output arrays. Note that setting squeeze_axis to $true$ removes axis with length 1 only along the axis which it is split. Also squeeze_axis can be set to $true$ only if $input.shape[axis] == num_outputs$.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.Softmax — Method.

Softmax(data, grad_scale, ignore_label, multi_output, use_ignore, preserve_shape, normalization, out_grad, smooth_alpha)

Please use SoftmaxOutput.

.. note::

This operator has been renamed to SoftmaxOutput, which computes the gradient of cross-entropy loss w.r.t softmax output. To just compute softmax output, use the softmax operator.

Defined in src/operator/softmax_output.cc:L138

Arguments

data::NDArray-or-SymbolicNode: Input array.
grad_scale::float, optional, default=1: Scales the gradient by a float factor.
ignore_label::float, optional, default=-1: The instances whose labels == ignore_label will be ignored during backward, if use_ignore is set to $true$).
multi_output::boolean, optional, default=0: If set to $true$, the softmax function will be computed along axis $1$. This is applied when the shape of input array differs from the shape of label array.
use_ignore::boolean, optional, default=0: If set to $true$, the ignore_label value will not contribute to the backward gradient.
preserve_shape::boolean, optional, default=0: If set to $true$, the softmax function will be computed along the last axis ($-1$).
normalization::{'batch', 'null', 'valid'},optional, default='null': Normalizes the gradient.
out_grad::boolean, optional, default=0: Multiplies gradient with output gradient element-wise.
smooth_alpha::float, optional, default=0: Constant for computing a label smoothed version of cross-entropyfor the backwards pass. This constant gets subtracted from theone-hot encoding of the gold label and distributed uniformly toall other labels.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.SoftmaxActivation — Method.

SoftmaxActivation(data, mode)

Applies softmax activation to input. This is intended for internal layers.

.. note::

This operator has been deprecated, please use softmax.

If mode = $instance$, this operator will compute a softmax for each instance in the batch. This is the default mode.

If mode = $channel$, this operator will compute a k-class softmax at each position of each instance, where k = $num_channel$. This mode can only be used when the input array has at least 3 dimensions. This can be used for fully convolutional network, image segmentation, etc.

Example::

input_array = mx.nd.array([[3., 0.5, -0.5, 2., 7.], [2., -.4, 7., 3., 0.2]]) softmax_act = mx.nd.SoftmaxActivation(input_array) print softmax_act.asnumpy()

[[ 1.78322066e-02 1.46375655e-03 5.38485940e-04 6.56010211e-03 9.73605454e-01] [ 6.56221947e-03 5.95310994e-04 9.73919690e-01 1.78379621e-02 1.08472735e-03]]

Defined in src/operator/nn/softmax_activation.cc:L67

Arguments

data::NDArray-or-SymbolicNode: Input array to activation function.
mode::{'channel', 'instance'},optional, default='instance': Specifies how to compute the softmax. If set to $instance$, it computes softmax for each instance. If set to $channel$, It computes cross channel softmax for each position of each instance.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.SoftmaxOutput — Method.

SoftmaxOutput(data, label, grad_scale, ignore_label, multi_output, use_ignore, preserve_shape, normalization, out_grad, smooth_alpha)

Computes the gradient of cross entropy loss with respect to softmax output.

This operator computes the gradient in two steps. The cross entropy loss does not actually need to be computed.
- Applies softmax function on the input array.
- Computes and returns the gradient of cross entropy loss w.r.t. the softmax output.
- The softmax function, cross entropy loss and gradient is given by:
- Softmax Function:
  
  .. math:: \text{softmax}(x)_i = \frac{exp(x_i)}{\sum_j exp(x_j)} * Cross Entropy Function:
  
  .. math:: \text{CE(label, output)} = - \sum_i \text{label}_i \log(\text{output}_i) * The gradient of cross entropy loss w.r.t softmax output:
  
  .. math:: \text{gradient} = \text{output} - \text{label} * During forward propagation, the softmax function is computed for each instance in the input array.
For general N-D input arrays with shape :math:(d_1, d_2, ..., d_n). The size is :math:s=d_1 \cdot d_2 \cdot \cdot \cdot d_n. We can use the parameters preserve_shape and multi_output to specify the way to compute softmax:
- By default, preserve_shape is $false$. This operator will reshape the input array into a 2-D array with shape :math:(d_1, \frac{s}{d_1}) and then compute the softmax function for each row in the reshaped array, and afterwards reshape it back to the original shape :math:(d_1, d_2, ..., d_n).
- If preserve_shape is $true$, the softmax function will be computed along the last axis (axis = $-1$).
- If multi_output is $true$, the softmax function will be computed along the second axis (axis = $1$).
- During backward propagation, the gradient of cross-entropy loss w.r.t softmax output array is computed. The provided label can be a one-hot label array or a probability label array.
- If the parameter use_ignore is $true$, ignore_label can specify input instances with a particular label to be ignored during backward propagation. This has no effect when softmax output has same shape as label.
  
  Example::
  
  data = [[1,2,3,4],[2,2,2,2],[3,3,3,3],[4,4,4,4]] label = [1,0,2,3] ignore_label = 1 SoftmaxOutput(data=data, label = label, multi_output=true, use_ignore=true, ignore_label=ignore_label)
  
  forward softmax output
  
  [[ 0.0320586 0.08714432 0.23688284 0.64391428] [ 0.25 0.25 0.25 0.25 ] [ 0.25 0.25 0.25 0.25 ] [ 0.25 0.25 0.25 0.25 ]]
  
  backward gradient output
  
  [[ 0. 0. 0. 0. ] [-0.75 0.25 0.25 0.25] [ 0.25 0.25 -0.75 0.25] [ 0.25 0.25 0.25 -0.75]]
  
  notice that the first row is all 0 because label[0] is 1, which is equal to ignore_label.
```
  * The parameter `grad_scale` can be used to rescale the gradient, which is often used to give each loss function different weights.
  * This operator also supports various ways to normalize the gradient by `normalization`, The `normalization` is applied if softmax output has different shape than the labels. The `normalization` mode can be set to the followings:
```
  - $'null'$: do nothing.
  - $'batch'$: divide the gradient by the batch size.
  - $'valid'$: divide the gradient by the number of instances which are not ignored.

Defined in src/operator/softmax_output.cc:L123

Arguments

data::NDArray-or-SymbolicNode: Input array.
label::NDArray-or-SymbolicNode: Ground truth label.
grad_scale::float, optional, default=1: Scales the gradient by a float factor.
ignore_label::float, optional, default=-1: The instances whose labels == ignore_label will be ignored during backward, if use_ignore is set to $true$).
multi_output::boolean, optional, default=0: If set to $true$, the softmax function will be computed along axis $1$. This is applied when the shape of input array differs from the shape of label array.
use_ignore::boolean, optional, default=0: If set to $true$, the ignore_label value will not contribute to the backward gradient.
preserve_shape::boolean, optional, default=0: If set to $true$, the softmax function will be computed along the last axis ($-1$).
normalization::{'batch', 'null', 'valid'},optional, default='null': Normalizes the gradient.
out_grad::boolean, optional, default=0: Multiplies gradient with output gradient element-wise.
smooth_alpha::float, optional, default=0: Constant for computing a label smoothed version of cross-entropyfor the backwards pass. This constant gets subtracted from theone-hot encoding of the gold label and distributed uniformly toall other labels.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.SpatialTransformer — Method.

SpatialTransformer(data, loc, target_shape, transform_type, sampler_type)

Applies a spatial transformer to input feature map.

Arguments

data::NDArray-or-SymbolicNode: Input data to the SpatialTransformerOp.
loc::NDArray-or-SymbolicNode: localisation net, the output dim should be 6 when transform_type is affine. You shold initialize the weight and bias with identity tranform.
target_shape::Shape(tuple), optional, default=[0,0]: output shape(h, w) of spatial transformer: (y, x)
transform_type::{'affine'}, required: transformation type
sampler_type::{'bilinear'}, required: sampling type
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.SwapAxis — Method.

SwapAxis(data, dim1, dim2)

Interchanges two axes of an array.

Examples::

x = [[1, 2, 3]]) swapaxes(x, 0, 1) = [[ 1], [ 2], [ 3]]

x = [[[ 0, 1], [ 2, 3]], [[ 4, 5], [ 6, 7]]] // (2,2,2) array

swapaxes(x, 0, 2) = [[[ 0, 4], [ 2, 6]], [[ 1, 5], [ 3, 7]]]

Defined in src/operator/swapaxis.cc:L70

Arguments

data::NDArray-or-SymbolicNode: Input array.
dim1::int (non-negative), optional, default=0: the first axis to be swapped.
dim2::int (non-negative), optional, default=0: the second axis to be swapped.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.UpSampling — Method.

UpSampling(data, scale, num_filter, sample_type, multi_input_mode, num_args, workspace)

Note: UpSampling takes variable number of positional inputs. So instead of calling as UpSampling([x, y, z], num_args=3), one should call via UpSampling(x, y, z), and num_args will be determined automatically.

Performs nearest neighbor/bilinear up sampling to inputs.

Arguments

data::NDArray-or-SymbolicNode[]: Array of tensors to upsample
scale::int (non-negative), required: Up sampling scale
num_filter::int (non-negative), optional, default=0: Input filter. Only used by bilinear sample_type.
sample_type::{'bilinear', 'nearest'}, required: upsampling method
multi_input_mode::{'concat', 'sum'},optional, default='concat': How to handle multiple input. concat means concatenate upsampled images along the channel dimension. sum means add all images together, only available for nearest neighbor upsampling.
num_args::int, required: Number of inputs to be upsampled. For nearest neighbor upsampling, this can be 1-N; the size of output will be(scaleh_0,scalew_0) and all other inputs will be upsampled to thesame size. For bilinear upsampling this must be 2; 1 input and 1 weight.
workspace::long (non-negative), optional, default=512: Tmp workspace for deconvolution (MB)
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._CachedOp — Method.

_CachedOp()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._CrossDeviceCopy — Method.

_CrossDeviceCopy()

Special op to copy data cross device

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._CustomFunction — Method.

_CustomFunction()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._Div — Method.

_Div(lhs, rhs)

_Div is an alias of elemwise_div.

Divides arguments element-wise.

The storage type of $elemwise_div$ output is always dense

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._DivScalar — Method.

_DivScalar(data, scalar)

_DivScalar is an alias of _div_scalar.

Divide an array with a scalar.

$_div_scalar$ only operates on data array of input if input is sparse.

For example, if input of shape (100, 100) has only 2 non zero elements, i.e. input.data = [5, 6], scalar = nan, it will result output.data = [nan, nan] instead of 10000 nans.

Defined in src/operator/tensor/elemwise_binary_scalar_op_basic.cc:L171

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._Equal — Method.

_Equal(lhs, rhs)

_Equal is an alias of _equal.

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._EqualScalar — Method.

_EqualScalar(data, scalar)

_EqualScalar is an alias of _equal_scalar.

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._Greater — Method.

_Greater(lhs, rhs)

_Greater is an alias of _greater.

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._GreaterEqualScalar — Method.

_GreaterEqualScalar(data, scalar)

_GreaterEqualScalar is an alias of _greater_equal_scalar.

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._GreaterScalar — Method.

_GreaterScalar(data, scalar)

_GreaterScalar is an alias of _greater_scalar.

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._Greater_Equal — Method.

_Greater_Equal(lhs, rhs)

_Greater_Equal is an alias of _greater_equal.

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._Hypot — Method.

_Hypot(lhs, rhs)

_Hypot is an alias of _hypot.

Given the "legs" of a right triangle, return its hypotenuse.

Defined in src/operator/tensor/elemwise_binary_op_extended.cc:L79

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._HypotScalar — Method.

_HypotScalar(data, scalar)

_HypotScalar is an alias of _hypot_scalar.

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._Lesser — Method.

_Lesser(lhs, rhs)

_Lesser is an alias of _lesser.

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._LesserEqualScalar — Method.

_LesserEqualScalar(data, scalar)

_LesserEqualScalar is an alias of _lesser_equal_scalar.

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._LesserScalar — Method.

_LesserScalar(data, scalar)

_LesserScalar is an alias of _lesser_scalar.

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._Lesser_Equal — Method.

_Lesser_Equal(lhs, rhs)

_Lesser_Equal is an alias of _lesser_equal.

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._Maximum — Method.

_Maximum(lhs, rhs)

_Maximum is an alias of _maximum.

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._MaximumScalar — Method.

_MaximumScalar(data, scalar)

_MaximumScalar is an alias of _maximum_scalar.

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._Minimum — Method.

_Minimum(lhs, rhs)

_Minimum is an alias of _minimum.

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._MinimumScalar — Method.

_MinimumScalar(data, scalar)

_MinimumScalar is an alias of _minimum_scalar.

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._Minus — Method.

_Minus(lhs, rhs)

_Minus is an alias of elemwise_sub.

Subtracts arguments element-wise.

The storage type of $elemwise_sub$ output depends on storage types of inputs

elemwise_sub(row_sparse, row_sparse) = row_sparse
elemwise_sub(csr, csr) = csr
otherwise, $elemwise_sub$ generates output with default storage

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._MinusScalar — Method.

_MinusScalar(data, scalar)

_MinusScalar is an alias of _minus_scalar.

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._Mod — Method.

_Mod(lhs, rhs)

_Mod is an alias of _mod.

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._ModScalar — Method.

_ModScalar(data, scalar)

_ModScalar is an alias of _mod_scalar.

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._Mul — Method.

_Mul(lhs, rhs)

_Mul is an alias of elemwise_mul.

Multiplies arguments element-wise.

The storage type of $elemwise_mul$ output depends on storage types of inputs

elemwise_mul(default, default) = default
elemwise_mul(row_sparse, row_sparse) = row_sparse
elemwise_mul(default, row_sparse) = default
elemwise_mul(row_sparse, default) = default
elemwise_mul(csr, csr) = csr
otherwise, $elemwise_mul$ generates output with default storage

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._MulScalar — Method.

_MulScalar(data, scalar)

_MulScalar is an alias of _mul_scalar.

Multiply an array with a scalar.

$_mul_scalar$ only operates on data array of input if input is sparse.

For example, if input of shape (100, 100) has only 2 non zero elements, i.e. input.data = [5, 6], scalar = nan, it will result output.data = [nan, nan] instead of 10000 nans.

Defined in src/operator/tensor/elemwise_binary_scalar_op_basic.cc:L149

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._NDArray — Method.

_NDArray(data, info)

Stub for implementing an operator implemented in native frontend language with ndarray.

Arguments

data::NDArray-or-SymbolicNode[]: Input data for the custom operator.
info::ptr, required:
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._Native — Method.

_Native(data, info, need_top_grad)

Stub for implementing an operator implemented in native frontend language.

Arguments

data::NDArray-or-SymbolicNode[]: Input data for the custom operator.
info::ptr, required:
need_top_grad::boolean, optional, default=1: Whether this layer needs out grad for backward. Should be false for loss layers.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._NoGradient — Method.

_NoGradient()

Place holder for variable who cannot perform gradient

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._NotEqualScalar — Method.

_NotEqualScalar(data, scalar)

_NotEqualScalar is an alias of _not_equal_scalar.

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._Not_Equal — Method.

_Not_Equal(lhs, rhs)

_Not_Equal is an alias of _not_equal.

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._Plus — Method.

_Plus(lhs, rhs)

_Plus is an alias of elemwise_add.

Adds arguments element-wise.

The storage type of $elemwise_add$ output depends on storage types of inputs

elemwise_add(row_sparse, row_sparse) = row_sparse
elemwise_add(csr, csr) = csr
otherwise, $elemwise_add$ generates output with default storage

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._PlusScalar — Method.

_PlusScalar(data, scalar)

_PlusScalar is an alias of _plus_scalar.

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._Power — Method.

_Power(lhs, rhs)

_Power is an alias of _power.

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._PowerScalar — Method.

_PowerScalar(data, scalar)

_PowerScalar is an alias of _power_scalar.

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._RDivScalar — Method.

_RDivScalar(data, scalar)

_RDivScalar is an alias of _rdiv_scalar.

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._RMinusScalar — Method.

_RMinusScalar(data, scalar)

_RMinusScalar is an alias of _rminus_scalar.

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._RModScalar — Method.

_RModScalar(data, scalar)

_RModScalar is an alias of _rmod_scalar.

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._RPowerScalar — Method.

_RPowerScalar(data, scalar)

_RPowerScalar is an alias of _rpower_scalar.

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._add — Method.

_add(lhs, rhs)

_add is an alias of elemwise_add.

Adds arguments element-wise.

The storage type of $elemwise_add$ output depends on storage types of inputs

elemwise_add(row_sparse, row_sparse) = row_sparse
elemwise_add(csr, csr) = csr
otherwise, $elemwise_add$ generates output with default storage

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._arange — Method.

_arange(start, stop, step, repeat, ctx, dtype)

Return evenly spaced values within a given interval. Similar to Numpy

Arguments

start::double, required: Start of interval. The interval includes this value. The default start value is 0.
stop::double or None, optional, default=None: End of interval. The interval does not include this value, except in some cases where step is not an integer and floating point round-off affects the length of out.
step::double, optional, default=1: Spacing between values.
repeat::int, optional, default='1': The repeating time of all elements. E.g repeat=3, the element a will be repeated three times –> a, a, a.
ctx::string, optional, default='': Context of output, in format cpu|gpu|cpu_pinned.Only used for imperative calls.
dtype::{'float16', 'float32', 'float64', 'int32', 'int64', 'uint8'},optional, default='float32': Target data type.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_Activation — Method.

_backward_Activation()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_BatchNorm — Method.

_backward_BatchNorm()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_BatchNorm_v1 — Method.

_backward_BatchNorm_v1()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_BilinearSampler — Method.

_backward_BilinearSampler()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_CachedOp — Method.

_backward_CachedOp()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_Concat — Method.

_backward_Concat()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_Convolution — Method.

_backward_Convolution()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_Convolution_v1 — Method.

_backward_Convolution_v1()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_Correlation — Method.

_backward_Correlation()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_Crop — Method.

_backward_Crop()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_Custom — Method.

_backward_Custom()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_CustomFunction — Method.

_backward_CustomFunction()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_Deconvolution — Method.

_backward_Deconvolution()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_Dropout — Method.

_backward_Dropout()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_Embedding — Method.

_backward_Embedding()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_FullyConnected — Method.

_backward_FullyConnected()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_GridGenerator — Method.

_backward_GridGenerator()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_IdentityAttachKLSparseReg — Method.

_backward_IdentityAttachKLSparseReg()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_InstanceNorm — Method.

_backward_InstanceNorm()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_L2Normalization — Method.

_backward_L2Normalization()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_LRN — Method.

_backward_LRN()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_LeakyReLU — Method.

_backward_LeakyReLU()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_MakeLoss — Method.

_backward_MakeLoss()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_Pad — Method.

_backward_Pad()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_Pooling — Method.

_backward_Pooling()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_Pooling_v1 — Method.

_backward_Pooling_v1()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_RNN — Method.

_backward_RNN()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_ROIPooling — Method.

_backward_ROIPooling()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_SVMOutput — Method.

_backward_SVMOutput()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_SequenceLast — Method.

_backward_SequenceLast()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_SequenceMask — Method.

_backward_SequenceMask()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_SequenceReverse — Method.

_backward_SequenceReverse()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_SliceChannel — Method.

_backward_SliceChannel()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_Softmax — Method.

_backward_Softmax()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_SoftmaxActivation — Method.

_backward_SoftmaxActivation()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_SoftmaxOutput — Method.

_backward_SoftmaxOutput()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_SparseEmbedding — Method.

_backward_SparseEmbedding()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_SpatialTransformer — Method.

_backward_SpatialTransformer()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_SwapAxis — Method.

_backward_SwapAxis()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_UpSampling — Method.

_backward_UpSampling()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward__CrossDeviceCopy — Method.

_backward__CrossDeviceCopy()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward__NDArray — Method.

_backward__NDArray()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward__Native — Method.

_backward__Native()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward__contrib_CTCLoss — Method.

_backward__contrib_CTCLoss()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward__contrib_DeformableConvolution — Method.

_backward__contrib_DeformableConvolution()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward__contrib_DeformablePSROIPooling — Method.

_backward__contrib_DeformablePSROIPooling()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward__contrib_MultiBoxDetection — Method.

_backward__contrib_MultiBoxDetection()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward__contrib_MultiBoxPrior — Method.

_backward__contrib_MultiBoxPrior()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward__contrib_MultiBoxTarget — Method.

_backward__contrib_MultiBoxTarget()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward__contrib_MultiProposal — Method.

_backward__contrib_MultiProposal()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward__contrib_PSROIPooling — Method.

_backward__contrib_PSROIPooling()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward__contrib_Proposal — Method.

_backward__contrib_Proposal()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward__contrib_count_sketch — Method.

_backward__contrib_count_sketch()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward__contrib_fft — Method.

_backward__contrib_fft()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward__contrib_ifft — Method.

_backward__contrib_ifft()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_abs — Method.

_backward_abs(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_add — Method.

_backward_add()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_arccos — Method.

_backward_arccos(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_arccosh — Method.

_backward_arccosh(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_arcsin — Method.

_backward_arcsin(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_arcsinh — Method.

_backward_arcsinh(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_arctan — Method.

_backward_arctan(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_arctanh — Method.

_backward_arctanh(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_batch_dot — Method.

_backward_batch_dot()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_broadcast_add — Method.

_backward_broadcast_add()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_broadcast_div — Method.

_backward_broadcast_div()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_broadcast_hypot — Method.

_backward_broadcast_hypot()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_broadcast_maximum — Method.

_backward_broadcast_maximum()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_broadcast_minimum — Method.

_backward_broadcast_minimum()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_broadcast_mod — Method.

_backward_broadcast_mod()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_broadcast_mul — Method.

_backward_broadcast_mul()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_broadcast_power — Method.

_backward_broadcast_power()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_broadcast_sub — Method.

_backward_broadcast_sub()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_cast — Method.

_backward_cast()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_cbrt — Method.

_backward_cbrt(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_clip — Method.

_backward_clip()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_contrib_bipartite_matching — Method.

_backward_contrib_bipartite_matching(is_ascend, threshold, topk)

Arguments

is_ascend::boolean, optional, default=0: Use ascend order for scores instead of descending. Please set threshold accordingly.
threshold::float, required: Ignore matching when score < thresh, if is_ascend=false, or ignore score > thresh, if is_ascend=true.
topk::int, optional, default='-1': Limit the number of matches to topk, set -1 for no limit
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_contrib_box_iou — Method.

_backward_contrib_box_iou(format)

Arguments

format::{'center', 'corner'},optional, default='corner': The box encoding type.

"corner" means boxes are encoded as [xmin, ymin, xmax, ymax], "center" means boxes are encodes as [x, y, width, height].

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_contrib_box_nms — Method.

_backward_contrib_box_nms(overlap_thresh, topk, coord_start, score_index, id_index, force_suppress, in_format, out_format)

Arguments

overlap_thresh::float, optional, default=0.5: Overlapping(IoU) threshold to suppress object with smaller score.
topk::int, optional, default='-1': Apply nms to topk boxes with descending scores, -1 to no restriction.
coord_start::int, optional, default='2': Start index of the consecutive 4 coordinates.
score_index::int, optional, default='1': Index of the scores/confidence of boxes.
id_index::int, optional, default='-1': Optional, index of the class categories, -1 to disable.
force_suppress::boolean, optional, default=0: Optional, if set false and id_index is provided, nms will only apply to boxes belongs to the same category
in_format::{'center', 'corner'},optional, default='corner': The input box encoding type.

"corner" means boxes are encoded as [xmin, ymin, xmax, ymax], "center" means boxes are encodes as [x, y, width, height].

out_format::{'center', 'corner'},optional, default='corner': The output box encoding type.

"corner" means boxes are encoded as [xmin, ymin, xmax, ymax], "center" means boxes are encodes as [x, y, width, height].

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_copy — Method.

_backward_copy()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_cos — Method.

_backward_cos(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_cosh — Method.

_backward_cosh(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_degrees — Method.

_backward_degrees(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_div — Method.

_backward_div()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_div_scalar — Method.

_backward_div_scalar(data, scalar)

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_dot — Method.

_backward_dot(transpose_a, transpose_b)

Arguments

transpose_a::boolean, optional, default=0: If true then transpose the first input before dot.
transpose_b::boolean, optional, default=0: If true then transpose the second input before dot.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_expm1 — Method.

_backward_expm1(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_gamma — Method.

_backward_gamma(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_gammaln — Method.

_backward_gammaln(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_gather_nd — Method.

_backward_gather_nd(data, indices, shape)

Accumulates data according to indices and get the result. It's the backward of gather_nd.

Given data with shape (Y_0, ..., Y_{K-1}, X_M, ..., X_{N-1}) and indices with shape (M, Y_0, ..., Y_{K-1}), the output will have shape (X_0, X_1, ..., X_{N-1}), where M <= N. If M == N, data shape should simply be (Y_0, ..., Y_{K-1}).

The elements in output is defined as follows::

output[indices[0, y_0, ..., y_{K-1}], ..., indices[M-1, y_0, ..., y_{K-1}], x_M, ..., x_{N-1}] += data[y_0, ..., y_{K-1}, x_M, ..., x_{N-1}]

all other entries in output are 0 or the original value if AddTo is triggered.

Examples::

data = [2, 3, 0] indices = [[1, 1, 0], [0, 1, 0]] shape = (2, 2) _backward_gather_nd(data, indices, shape) = [[0, 0], [2, 3]] # Same as scatter_nd

The difference between scatter_nd and scatter_nd_acc is the latter will accumulate

the values that point to the same index.

data = [2, 3, 0] indices = [[1, 1, 0], [1, 1, 0]] shape = (2, 2) _backward_gather_nd(data, indices, shape) = [[0, 0], [0, 5]]

Arguments

data::NDArray-or-SymbolicNode: data
indices::NDArray-or-SymbolicNode: indices
shape::Shape(tuple), required: Shape of output.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_hypot — Method.

_backward_hypot()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_hypot_scalar — Method.

_backward_hypot_scalar(lhs, rhs, scalar)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
scalar::float: scalar value
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_linalg_gelqf — Method.

_backward_linalg_gelqf()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_linalg_gemm — Method.

_backward_linalg_gemm()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_linalg_gemm2 — Method.

_backward_linalg_gemm2()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_linalg_potrf — Method.

_backward_linalg_potrf()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_linalg_potri — Method.

_backward_linalg_potri()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_linalg_sumlogdiag — Method.

_backward_linalg_sumlogdiag()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_linalg_syevd — Method.

_backward_linalg_syevd()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_linalg_syrk — Method.

_backward_linalg_syrk()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_linalg_trmm — Method.

_backward_linalg_trmm()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_linalg_trsm — Method.

_backward_linalg_trsm()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_linear_reg_out — Method.

_backward_linear_reg_out()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_log — Method.

_backward_log(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_log10 — Method.

_backward_log10(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_log1p — Method.

_backward_log1p(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_log2 — Method.

_backward_log2(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_log_softmax — Method.

_backward_log_softmax(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_logistic_reg_out — Method.

_backward_logistic_reg_out()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_mae_reg_out — Method.

_backward_mae_reg_out()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_max — Method.

_backward_max()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_maximum — Method.

_backward_maximum()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_maximum_scalar — Method.

_backward_maximum_scalar(lhs, rhs, scalar)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
scalar::float: scalar value
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_mean — Method.

_backward_mean()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_min — Method.

_backward_min()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_minimum — Method.

_backward_minimum()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_minimum_scalar — Method.

_backward_minimum_scalar(lhs, rhs, scalar)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
scalar::float: scalar value
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_mod — Method.

_backward_mod()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_mod_scalar — Method.

_backward_mod_scalar(lhs, rhs, scalar)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
scalar::float: scalar value
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_mul — Method.

_backward_mul()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_mul_scalar — Method.

_backward_mul_scalar(data, scalar)

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_nanprod — Method.

_backward_nanprod()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_nansum — Method.

_backward_nansum()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_pick — Method.

_backward_pick()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_power — Method.

_backward_power()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_power_scalar — Method.

_backward_power_scalar(lhs, rhs, scalar)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
scalar::float: scalar value
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_prod — Method.

_backward_prod()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_radians — Method.

_backward_radians(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_rcbrt — Method.

_backward_rcbrt(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_rdiv_scalar — Method.

_backward_rdiv_scalar(lhs, rhs, scalar)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
scalar::float: scalar value
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_reciprocal — Method.

_backward_reciprocal(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_relu — Method.

_backward_relu(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_repeat — Method.

_backward_repeat()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_reverse — Method.

_backward_reverse()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_rmod_scalar — Method.

_backward_rmod_scalar(lhs, rhs, scalar)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
scalar::float: scalar value
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_rpower_scalar — Method.

_backward_rpower_scalar(lhs, rhs, scalar)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
scalar::float: scalar value
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_rsqrt — Method.

_backward_rsqrt(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_sample_multinomial — Method.

_backward_sample_multinomial()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_sigmoid — Method.

_backward_sigmoid(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_sign — Method.

_backward_sign(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_sin — Method.

_backward_sin(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_sinh — Method.

_backward_sinh(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_slice — Method.

_backward_slice()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_slice_axis — Method.

_backward_slice_axis()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_smooth_l1 — Method.

_backward_smooth_l1(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_softmax — Method.

_backward_softmax(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_softmax_cross_entropy — Method.

_backward_softmax_cross_entropy()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_sparse_retain — Method.

_backward_sparse_retain()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_sqrt — Method.

_backward_sqrt(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_square — Method.

_backward_square(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_square_sum — Method.

_backward_square_sum()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_squeeze — Method.

_backward_squeeze()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_stack — Method.

_backward_stack()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_sub — Method.

_backward_sub()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_sum — Method.

_backward_sum()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_take — Method.

_backward_take()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_tan — Method.

_backward_tan(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_tanh — Method.

_backward_tanh(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_tile — Method.

_backward_tile()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_topk — Method.

_backward_topk()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._backward_where — Method.

_backward_where()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._broadcast_backward — Method.

_broadcast_backward()

Arguments

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._contrib_CTCLoss — Method.

_contrib_CTCLoss(data, label, data_lengths, label_lengths, use_data_lengths, use_label_lengths, blank_label)

Connectionist Temporal Classification Loss.

The shapes of the inputs and outputs:

data: (sequence_length, batch_size, alphabet_size)
label: (batch_size, label_sequence_length)
out: (batch_size)

The data tensor consists of sequences of activation vectors (without applying softmax), with i-th channel in the last dimension corresponding to i-th label for i between 0 and alphabet_size-1 (i.e always 0-indexed). Alphabet size should include one additional value reserved for blank label. When blank_label is $"first"$, the $0$-th channel is be reserved for activation of blank label, or otherwise if it is "last", $(alphabet_size-1)$-th channel should be reserved for blank label.

$label$ is an index matrix of integers. When blank_label is $"first"$, the value 0 is then reserved for blank label, and should not be passed in this matrix. Otherwise, when blank_label is $"last"$, the value (alphabet_size-1) is reserved for blank label.

If a sequence of labels is shorter than label_sequence_length, use the special padding value at the end of the sequence to conform it to the correct length. The padding value is 0 when blank_label is $"first"$, and -1 otherwise.

For example, suppose the vocabulary is [a, b, c], and in one batch we have three sequences 'ba', 'cbb', and 'abac'. When blank_label is $"first"$, we can index the labels as {'a': 1, 'b': 2, 'c': 3}, and we reserve the 0-th channel for blank label in data tensor. The resulting label tensor should be padded to be::

[[2, 1, 0, 0], [3, 2, 2, 0], [1, 2, 1, 3]]

When blank_label is $"last"$, we can index the labels as {'a': 0, 'b': 1, 'c': 2}, and we reserve the channel index 3 for blank label in data tensor. The resulting label tensor should be padded to be::

[[1, 0, -1, -1], [2, 1, 1, -1], [0, 1, 0, 2]]

$out$ is a list of CTC loss values, one per example in the batch.

See Connectionist Temporal Classification: Labelling Unsegmented Sequence Data with Recurrent Neural Networks, A. Graves et al. for more information on the definition and the algorithm.

Defined in src/operator/contrib/ctc_loss.cc:L115

Arguments

data::NDArray-or-SymbolicNode: Input data to the ctc_loss op.
label::NDArray-or-SymbolicNode: Ground-truth labels for the loss.
data_lengths::NDArray-or-SymbolicNode: Lengths of data for each of the samples. Only required when use_data_lengths is true.
label_lengths::NDArray-or-SymbolicNode: Lengths of labels for each of the samples. Only required when use_label_lengths is true.
use_data_lengths::boolean, optional, default=0: Whether the data lenghts are decided by data_lengths. If false, the lengths are equal to the max sequence length.
use_label_lengths::boolean, optional, default=0: Whether the label lenghts are decided by label_lengths, or derived from padding_mask. If false, the lengths are derived from the first occurrence of the value of padding_mask. The value of padding_mask is $0$ when first CTC label is reserved for blank, and $-1$ when last label is reserved for blank. See blank_label.
blank_label::{'first', 'last'},optional, default='first': Set the label that is reserved for blank label.If "first", 0-th label is reserved, and label values for tokens in the vocabulary are between $1$ and $alphabet_size-1$, and the padding mask is $-1$. If "last", last label value $alphabet_size-1$ is reserved for blank label instead, and label values for tokens in the vocabulary are between $0$ and $alphabet_size-2$, and the padding mask is $0$.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._contrib_DeformableConvolution — Method.

_contrib_DeformableConvolution(data, offset, weight, bias, kernel, stride, dilate, pad, num_filter, num_group, num_deformable_group, workspace, no_bias, layout)

Compute 2-D deformable convolution on 4-D input.

The deformable convolution operation is described in https://arxiv.org/abs/1703.06211

For 2-D deformable convolution, the shapes are

data: (batch_size, channel, height, width)
offset: (batch_size, num_deformable_group * kernel[0] * kernel[1], height, width)
weight: (num_filter, channel, kernel[0], kernel[1])
bias: (num_filter,)
out: (batch_size, num_filter, out_height, out_width).

Define::

f(x,k,p,s,d) = floor((x+2p-d(k-1)-1)/s)+1

then we have::

out_height=f(height, kernel[0], pad[0], stride[0], dilate[0]) out_width=f(width, kernel[1], pad[1], stride[1], dilate[1])

If $no_bias$ is set to be true, then the $bias$ term is ignored.

The default data $layout$ is NCHW, namely (batch_size, channle, height, width).

If $num_group$ is larger than 1, denoted by g, then split the input $data$ evenly into g parts along the channel axis, and also evenly split $weight$ along the first dimension. Next compute the convolution on the i-th part of the data with the i-th weight part. The output is obtained by concating all the g results.

If $num_deformable_group$ is larger than 1, denoted by dg, then split the input $offset$ evenly into dg parts along the channel axis, and also evenly split $out$ evenly into dg parts along the channel axis. Next compute the deformable convolution, apply the i-th part of the offset part on the i-th out.

Both $weight$ and $bias$ are learnable parameters.

Defined in src/operator/contrib/deformable_convolution.cc:L100

Arguments

data::NDArray-or-SymbolicNode: Input data to the DeformableConvolutionOp.
offset::NDArray-or-SymbolicNode: Input offset to the DeformableConvolutionOp.
weight::NDArray-or-SymbolicNode: Weight matrix.
bias::NDArray-or-SymbolicNode: Bias parameter.
kernel::Shape(tuple), required: Convolution kernel size: (h, w) or (d, h, w)
stride::Shape(tuple), optional, default=[]: Convolution stride: (h, w) or (d, h, w). Defaults to 1 for each dimension.
dilate::Shape(tuple), optional, default=[]: Convolution dilate: (h, w) or (d, h, w). Defaults to 1 for each dimension.
pad::Shape(tuple), optional, default=[]: Zero pad for convolution: (h, w) or (d, h, w). Defaults to no padding.
num_filter::int (non-negative), required: Convolution filter(channel) number
num_group::int (non-negative), optional, default=1: Number of group partitions.
num_deformable_group::int (non-negative), optional, default=1: Number of deformable group partitions.
workspace::long (non-negative), optional, default=1024: Maximum temperal workspace allowed for convolution (MB).
no_bias::boolean, optional, default=0: Whether to disable bias parameter.
layout::{None, 'NCDHW', 'NCHW', 'NCW'},optional, default='None': Set layout for input, output and weight. Empty for default layout: NCW for 1d, NCHW for 2d and NCDHW for 3d.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._contrib_DeformablePSROIPooling — Method.

_contrib_DeformablePSROIPooling(data, rois, trans, spatial_scale, output_dim, group_size, pooled_size, part_size, sample_per_part, trans_std, no_trans)

Performs deformable position-sensitive region-of-interest pooling on inputs. The DeformablePSROIPooling operation is described in https://arxiv.org/abs/1703.06211 .batch_size will change to the number of region bounding boxes after DeformablePSROIPooling

Arguments

data::SymbolicNode: Input data to the pooling operator, a 4D Feature maps
rois::SymbolicNode: Bounding box coordinates, a 2D array of [[batch_index, x1, y1, x2, y2]]. (x1, y1) and (x2, y2) are top left and down right corners of designated region of interest. batch_index indicates the index of corresponding image in the input data
trans::SymbolicNode: transition parameter
spatial_scale::float, required: Ratio of input feature map height (or w) to raw image height (or w). Equals the reciprocal of total stride in convolutional layers
output_dim::int, required: fix output dim
group_size::int, required: fix group size
pooled_size::int, required: fix pooled size
part_size::int, optional, default='0': fix part size
sample_per_part::int, optional, default='1': fix samples per part
trans_std::float, optional, default=0: fix transition std
no_trans::boolean, optional, default=0: Whether to disable trans parameter.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._contrib_MultiBoxDetection — Method.

_contrib_MultiBoxDetection(cls_prob, loc_pred, anchor, clip, threshold, background_id, nms_threshold, force_suppress, variances, nms_topk)

Convert multibox detection predictions.

Arguments

cls_prob::NDArray-or-SymbolicNode: Class probabilities.
loc_pred::NDArray-or-SymbolicNode: Location regression predictions.
anchor::NDArray-or-SymbolicNode: Multibox prior anchor boxes
clip::boolean, optional, default=1: Clip out-of-boundary boxes.
threshold::float, optional, default=0.01: Threshold to be a positive prediction.
background_id::int, optional, default='0': Background id.
nms_threshold::float, optional, default=0.5: Non-maximum suppression threshold.
force_suppress::boolean, optional, default=0: Suppress all detections regardless of class_id.
variances::tuple of <float>, optional, default=[0.1,0.1,0.2,0.2]: Variances to be decoded from box regression output.
nms_topk::int, optional, default='-1': Keep maximum top k detections before nms, -1 for no limit.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._contrib_MultiBoxPrior — Method.

_contrib_MultiBoxPrior(data, sizes, ratios, clip, steps, offsets)

Generate prior(anchor) boxes from data, sizes and ratios.

Arguments

data::NDArray-or-SymbolicNode: Input data.
sizes::tuple of <float>, optional, default=[1]: List of sizes of generated MultiBoxPriores.
ratios::tuple of <float>, optional, default=[1]: List of aspect ratios of generated MultiBoxPriores.
clip::boolean, optional, default=0: Whether to clip out-of-boundary boxes.
steps::tuple of <float>, optional, default=[-1,-1]: Priorbox step across y and x, -1 for auto calculation.
offsets::tuple of <float>, optional, default=[0.5,0.5]: Priorbox center offsets, y and x respectively
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._contrib_MultiBoxTarget — Method.

_contrib_MultiBoxTarget(anchor, label, cls_pred, overlap_threshold, ignore_label, negative_mining_ratio, negative_mining_thresh, minimum_negative_samples, variances)

Compute Multibox training targets

Arguments

anchor::NDArray-or-SymbolicNode: Generated anchor boxes.
label::NDArray-or-SymbolicNode: Object detection labels.
cls_pred::NDArray-or-SymbolicNode: Class predictions.
overlap_threshold::float, optional, default=0.5: Anchor-GT overlap threshold to be regarded as a positive match.
ignore_label::float, optional, default=-1: Label for ignored anchors.
negative_mining_ratio::float, optional, default=-1: Max negative to positive samples ratio, use -1 to disable mining
negative_mining_thresh::float, optional, default=0.5: Threshold used for negative mining.
minimum_negative_samples::int, optional, default='0': Minimum number of negative samples.
variances::tuple of <float>, optional, default=[0.1,0.1,0.2,0.2]: Variances to be encoded in box regression target.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._contrib_MultiProposal — Method.

_contrib_MultiProposal(cls_score, bbox_pred, im_info, rpn_pre_nms_top_n, rpn_post_nms_top_n, threshold, rpn_min_size, scales, ratios, feature_stride, output_score, iou_loss)

Generate region proposals via RPN

Arguments

cls_score::NDArray-or-SymbolicNode: Score of how likely proposal is object.
bbox_pred::NDArray-or-SymbolicNode: BBox Predicted deltas from anchors for proposals
im_info::NDArray-or-SymbolicNode: Image size and scale.
rpn_pre_nms_top_n::int, optional, default='6000': Number of top scoring boxes to keep after applying NMS to RPN proposals
rpn_post_nms_top_n::int, optional, default='300': Overlap threshold used for non-maximumsuppresion(suppress boxes with IoU >= this threshold
threshold::float, optional, default=0.7: NMS value, below which to suppress.
rpn_min_size::int, optional, default='16': Minimum height or width in proposal
scales::tuple of <float>, optional, default=[4,8,16,32]: Used to generate anchor windows by enumerating scales
ratios::tuple of <float>, optional, default=[0.5,1,2]: Used to generate anchor windows by enumerating ratios
feature_stride::int, optional, default='16': The size of the receptive field each unit in the convolution layer of the rpn,for example the product of all stride's prior to this layer.
output_score::boolean, optional, default=0: Add score to outputs
iou_loss::boolean, optional, default=0: Usage of IoU Loss
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._contrib_PSROIPooling — Method.

_contrib_PSROIPooling(data, rois, spatial_scale, output_dim, pooled_size, group_size)

Performs region-of-interest pooling on inputs. Resize bounding box coordinates by spatial_scale and crop input feature maps accordingly. The cropped feature maps are pooled by max pooling to a fixed size output indicated by pooled_size. batch_size will change to the number of region bounding boxes after PSROIPooling

Arguments

data::SymbolicNode: Input data to the pooling operator, a 4D Feature maps
rois::SymbolicNode: Bounding box coordinates, a 2D array of [[batch_index, x1, y1, x2, y2]]. (x1, y1) and (x2, y2) are top left and down right corners of designated region of interest. batch_index indicates the index of corresponding image in the input data
spatial_scale::float, required: Ratio of input feature map height (or w) to raw image height (or w). Equals the reciprocal of total stride in convolutional layers
output_dim::int, required: fix output dim
pooled_size::int, required: fix pooled size
group_size::int, optional, default='0': fix group size
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._contrib_Proposal — Method.

_contrib_Proposal(cls_score, bbox_pred, im_info, rpn_pre_nms_top_n, rpn_post_nms_top_n, threshold, rpn_min_size, scales, ratios, feature_stride, output_score, iou_loss)

Generate region proposals via RPN

Arguments

cls_score::NDArray-or-SymbolicNode: Score of how likely proposal is object.
bbox_pred::NDArray-or-SymbolicNode: BBox Predicted deltas from anchors for proposals
im_info::NDArray-or-SymbolicNode: Image size and scale.
rpn_pre_nms_top_n::int, optional, default='6000': Number of top scoring boxes to keep after applying NMS to RPN proposals
rpn_post_nms_top_n::int, optional, default='300': Overlap threshold used for non-maximumsuppresion(suppress boxes with IoU >= this threshold
threshold::float, optional, default=0.7: NMS value, below which to suppress.
rpn_min_size::int, optional, default='16': Minimum height or width in proposal
scales::tuple of <float>, optional, default=[4,8,16,32]: Used to generate anchor windows by enumerating scales
ratios::tuple of <float>, optional, default=[0.5,1,2]: Used to generate anchor windows by enumerating ratios
feature_stride::int, optional, default='16': The size of the receptive field each unit in the convolution layer of the rpn,for example the product of all stride's prior to this layer.
output_score::boolean, optional, default=0: Add score to outputs
iou_loss::boolean, optional, default=0: Usage of IoU Loss
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._contrib_SparseEmbedding — Method.

_contrib_SparseEmbedding(data, weight, input_dim, output_dim, dtype)

Maps integer indices to vector representations (embeddings).

This operator maps words to real-valued vectors in a high-dimensional space, called word embeddings. These embeddings can capture semantic and syntactic properties of the words. For example, it has been noted that in the learned embedding spaces, similar words tend to be close to each other and dissimilar words far apart.

For an input array of shape (d1, ..., dK), the shape of an output array is (d1, ..., dK, output_dim). All the input values should be integers in the range [0, input_dim).

If the input_dim is ip0 and output_dim is op0, then shape of the embedding weight matrix must be (ip0, op0).

The storage type of weight must be row_sparse, and the gradient of the weight will be of row_sparse storage type, too.

.. Note::

`SparseEmbedding` is designed for the use case where `input_dim` is very large (e.g. 100k).
The operator is available on both CPU and GPU.

Examples::

input_dim = 4 output_dim = 5

// Each row in weight matrix y represents a word. So, y = (w0,w1,w2,w3) y = [[ 0., 1., 2., 3., 4.], [ 5., 6., 7., 8., 9.], [ 10., 11., 12., 13., 14.], [ 15., 16., 17., 18., 19.]]

// Input array x represents n-grams(2-gram). So, x = [(w1,w3), (w0,w2)] x = [[ 1., 3.], [ 0., 2.]]

// Mapped input x to its vector representation y. SparseEmbedding(x, y, 4, 5) = [[[ 5., 6., 7., 8., 9.], [ 15., 16., 17., 18., 19.]],

                            [[  0.,   1.,   2.,   3.,   4.],
                             [ 10.,  11.,  12.,  13.,  14.]]]

Defined in src/operator/tensor/indexing_op.cc:L294

Arguments

data::NDArray-or-SymbolicNode: The input array to the embedding operator.
weight::NDArray-or-SymbolicNode: The embedding weight matrix.
input_dim::int, required: Vocabulary size of the input indices.
output_dim::int, required: Dimension of the embedding vectors.
dtype::{'float16', 'float32', 'float64', 'int32', 'uint8'},optional, default='float32': Data type of weight.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._contrib_bipartite_matching — Method.

_contrib_bipartite_matching(data, is_ascend, threshold, topk)

Compute bipartite matching. The matching is performed on score matrix with shape [B, N, M]

B: batch_size
N: number of rows to match
M: number of columns as reference to be matched against.

Returns: x : matched column indices. -1 indicating non-matched elements in rows. y : matched row indices.

Note::

Zero gradients are back-propagated in this op for now.

Example::

s = [[0.5, 0.6], [0.1, 0.2], [0.3, 0.4]]
x, y = bipartite_matching(x, threshold=1e-12, is_ascend=False)
x = [1, -1, 0]
y = [2, 0]

Defined in src/operator/contrib/bounding_box.cc:L169

Arguments

data::NDArray-or-SymbolicNode: The input
is_ascend::boolean, optional, default=0: Use ascend order for scores instead of descending. Please set threshold accordingly.
threshold::float, required: Ignore matching when score < thresh, if is_ascend=false, or ignore score > thresh, if is_ascend=true.
topk::int, optional, default='-1': Limit the number of matches to topk, set -1 for no limit
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._contrib_box_iou — Method.

_contrib_box_iou(lhs, rhs, format)

Bounding box overlap of two arrays. The overlap is defined as Intersection-over-Union, aka, IOU.

lhs: (a_1, a_2, ..., a_n, 4) array
rhs: (b_1, b_2, ..., b_n, 4) array
output: (a_1, a_2, ..., a_n, b_1, b_2, ..., b_n) array

Note::

Zero gradients are back-propagated in this op for now.

Example::

x = [[0.5, 0.5, 1.0, 1.0], [0.0, 0.0, 0.5, 0.5]]
y = [0.25, 0.25, 0.75, 0.75]
box_iou(x, y, format='corner') = [[0.1428], [0.1428]]

Defined in src/operator/contrib/bounding_box.cc:L123

Arguments

lhs::NDArray-or-SymbolicNode: The first input
rhs::NDArray-or-SymbolicNode: The second input
format::{'center', 'corner'},optional, default='corner': The box encoding type.

"corner" means boxes are encoded as [xmin, ymin, xmax, ymax], "center" means boxes are encodes as [x, y, width, height].

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._contrib_box_nms — Method.

_contrib_box_nms(data, overlap_thresh, topk, coord_start, score_index, id_index, force_suppress, in_format, out_format)

Apply non-maximum suppression to input.

The output will be sorted in descending order according to score. Boxes with overlaps larger than overlap_thresh and smaller scores will be removed and filled with -1, the corresponding position will be recorded for backward propogation.

During back-propagation, the gradient will be copied to the original position according to the input index. For positions that have been suppressed, the in_grad will be assigned 0. In summary, gradients are sticked to its boxes, will either be moved or discarded according to its original index in input.

Input requirements:

Input tensor have at least 2 dimensions, (n, k), any higher dims will be regarded

as batch, e.g. (a, b, c, d, n, k) == (abc*d, n, k)

n is the number of boxes in each batch
k is the width of each box item.

By default, a box is [id, score, xmin, ymin, xmax, ymax, ...], additional elements are allowed.

id_index: optional, use -1 to ignore, useful if force_suppress=False, which means

we will skip highly overlapped boxes if one is apple while the other is car.

coord_start: required, default=2, the starting index of the 4 coordinates.

Two formats are supported: corner: [xmin, ymin, xmax, ymax] center: [x, y, width, height]

score_index: required, default=1, box score/confidence.

When two boxes overlap IOU > overlap_thresh, the one with smaller score will be suppressed.

in_format and out_format: default='corner', specify in/out box formats.

Examples::

x = [[0, 0.5, 0.1, 0.1, 0.2, 0.2], [1, 0.4, 0.1, 0.1, 0.2, 0.2], [0, 0.3, 0.1, 0.1, 0.14, 0.14], [2, 0.6, 0.5, 0.5, 0.7, 0.8]] box_nms(x, overlap_thresh=0.1, coord_start=2, score_index=1, id_index=0, force_suppress=True, in_format='corner', out_typ='corner') = [[2, 0.6, 0.5, 0.5, 0.7, 0.8], [0, 0.5, 0.1, 0.1, 0.2, 0.2], [-1, -1, -1, -1, -1, -1], [-1, -1, -1, -1, -1, -1]] out_grad = [[0.1, 0.1, 0.1, 0.1, 0.1, 0.1], [0.2, 0.2, 0.2, 0.2, 0.2, 0.2], [0.3, 0.3, 0.3, 0.3, 0.3, 0.3], [0.4, 0.4, 0.4, 0.4, 0.4, 0.4]]

exe.backward

in_grad = [[0.2, 0.2, 0.2, 0.2, 0.2, 0.2], [0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0], [0.1, 0.1, 0.1, 0.1, 0.1, 0.1]]

Defined in src/operator/contrib/bounding_box.cc:L82

Arguments

data::NDArray-or-SymbolicNode: The input
overlap_thresh::float, optional, default=0.5: Overlapping(IoU) threshold to suppress object with smaller score.
topk::int, optional, default='-1': Apply nms to topk boxes with descending scores, -1 to no restriction.
coord_start::int, optional, default='2': Start index of the consecutive 4 coordinates.
score_index::int, optional, default='1': Index of the scores/confidence of boxes.
id_index::int, optional, default='-1': Optional, index of the class categories, -1 to disable.
force_suppress::boolean, optional, default=0: Optional, if set false and id_index is provided, nms will only apply to boxes belongs to the same category
in_format::{'center', 'corner'},optional, default='corner': The input box encoding type.

"corner" means boxes are encoded as [xmin, ymin, xmax, ymax], "center" means boxes are encodes as [x, y, width, height].

out_format::{'center', 'corner'},optional, default='corner': The output box encoding type.

"corner" means boxes are encoded as [xmin, ymin, xmax, ymax], "center" means boxes are encodes as [x, y, width, height].

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._contrib_box_non_maximum_suppression — Method.

_contrib_box_non_maximum_suppression(data, overlap_thresh, topk, coord_start, score_index, id_index, force_suppress, in_format, out_format)

_contrib_box_non_maximum_suppression is an alias of _contrib_box_nms.

Apply non-maximum suppression to input.

The output will be sorted in descending order according to score. Boxes with overlaps larger than overlap_thresh and smaller scores will be removed and filled with -1, the corresponding position will be recorded for backward propogation.

During back-propagation, the gradient will be copied to the original position according to the input index. For positions that have been suppressed, the in_grad will be assigned 0. In summary, gradients are sticked to its boxes, will either be moved or discarded according to its original index in input.

Input requirements:

Input tensor have at least 2 dimensions, (n, k), any higher dims will be regarded

as batch, e.g. (a, b, c, d, n, k) == (abc*d, n, k)

n is the number of boxes in each batch
k is the width of each box item.

By default, a box is [id, score, xmin, ymin, xmax, ymax, ...], additional elements are allowed.

id_index: optional, use -1 to ignore, useful if force_suppress=False, which means

we will skip highly overlapped boxes if one is apple while the other is car.

coord_start: required, default=2, the starting index of the 4 coordinates.

Two formats are supported: corner: [xmin, ymin, xmax, ymax] center: [x, y, width, height]

score_index: required, default=1, box score/confidence.

When two boxes overlap IOU > overlap_thresh, the one with smaller score will be suppressed.

in_format and out_format: default='corner', specify in/out box formats.

Examples::

x = [[0, 0.5, 0.1, 0.1, 0.2, 0.2], [1, 0.4, 0.1, 0.1, 0.2, 0.2], [0, 0.3, 0.1, 0.1, 0.14, 0.14], [2, 0.6, 0.5, 0.5, 0.7, 0.8]] box_nms(x, overlap_thresh=0.1, coord_start=2, score_index=1, id_index=0, force_suppress=True, in_format='corner', out_typ='corner') = [[2, 0.6, 0.5, 0.5, 0.7, 0.8], [0, 0.5, 0.1, 0.1, 0.2, 0.2], [-1, -1, -1, -1, -1, -1], [-1, -1, -1, -1, -1, -1]] out_grad = [[0.1, 0.1, 0.1, 0.1, 0.1, 0.1], [0.2, 0.2, 0.2, 0.2, 0.2, 0.2], [0.3, 0.3, 0.3, 0.3, 0.3, 0.3], [0.4, 0.4, 0.4, 0.4, 0.4, 0.4]]

exe.backward

in_grad = [[0.2, 0.2, 0.2, 0.2, 0.2, 0.2], [0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0], [0.1, 0.1, 0.1, 0.1, 0.1, 0.1]]

Defined in src/operator/contrib/bounding_box.cc:L82

Arguments

data::NDArray-or-SymbolicNode: The input
overlap_thresh::float, optional, default=0.5: Overlapping(IoU) threshold to suppress object with smaller score.
topk::int, optional, default='-1': Apply nms to topk boxes with descending scores, -1 to no restriction.
coord_start::int, optional, default='2': Start index of the consecutive 4 coordinates.
score_index::int, optional, default='1': Index of the scores/confidence of boxes.
id_index::int, optional, default='-1': Optional, index of the class categories, -1 to disable.
force_suppress::boolean, optional, default=0: Optional, if set false and id_index is provided, nms will only apply to boxes belongs to the same category
in_format::{'center', 'corner'},optional, default='corner': The input box encoding type.

"corner" means boxes are encoded as [xmin, ymin, xmax, ymax], "center" means boxes are encodes as [x, y, width, height].

out_format::{'center', 'corner'},optional, default='corner': The output box encoding type.

"corner" means boxes are encoded as [xmin, ymin, xmax, ymax], "center" means boxes are encodes as [x, y, width, height].

name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._contrib_count_sketch — Method.

_contrib_count_sketch(data, h, s, out_dim, processing_batch_size)

Apply CountSketch to input: map a d-dimension data to k-dimension data"

.. note:: count_sketch is only available on GPU.

Assume input data has shape (N, d), sign hash table s has shape (N, d), index hash table h has shape (N, d) and mapping dimension out_dim = k, each element in s is either +1 or -1, each element in h is random integer from 0 to k-1. Then the operator computs:

.. math:: out[h[i]] += data[i] * s[i]

Example::

out_dim = 5 x = [[1.2, 2.5, 3.4],[3.2, 5.7, 6.6]] h = [[0, 3, 4]] s = [[1, -1, 1]] mx.contrib.ndarray.count_sketch(data=x, h=h, s=s, out_dim = 5) = [[1.2, 0, 0, -2.5, 3.4], [3.2, 0, 0, -5.7, 6.6]]

Defined in src/operator/contrib/count_sketch.cc:L67

Arguments

data::NDArray-or-SymbolicNode: Input data to the CountSketchOp.
h::NDArray-or-SymbolicNode: The index vector
s::NDArray-or-SymbolicNode: The sign vector
out_dim::int, required: The output dimension.
processing_batch_size::int, optional, default='32': How many sketch vectors to process at one time.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._contrib_ctc_loss — Method.

_contrib_ctc_loss(data, label, data_lengths, label_lengths, use_data_lengths, use_label_lengths, blank_label)

_contrib_ctc_loss is an alias of _contrib_CTCLoss.

Connectionist Temporal Classification Loss.

The shapes of the inputs and outputs:

data: (sequence_length, batch_size, alphabet_size)
label: (batch_size, label_sequence_length)
out: (batch_size)

The data tensor consists of sequences of activation vectors (without applying softmax), with i-th channel in the last dimension corresponding to i-th label for i between 0 and alphabet_size-1 (i.e always 0-indexed). Alphabet size should include one additional value reserved for blank label. When blank_label is $"first"$, the $0$-th channel is be reserved for activation of blank label, or otherwise if it is "last", $(alphabet_size-1)$-th channel should be reserved for blank label.

$label$ is an index matrix of integers. When blank_label is $"first"$, the value 0 is then reserved for blank label, and should not be passed in this matrix. Otherwise, when blank_label is $"last"$, the value (alphabet_size-1) is reserved for blank label.

If a sequence of labels is shorter than label_sequence_length, use the special padding value at the end of the sequence to conform it to the correct length. The padding value is 0 when blank_label is $"first"$, and -1 otherwise.

For example, suppose the vocabulary is [a, b, c], and in one batch we have three sequences 'ba', 'cbb', and 'abac'. When blank_label is $"first"$, we can index the labels as {'a': 1, 'b': 2, 'c': 3}, and we reserve the 0-th channel for blank label in data tensor. The resulting label tensor should be padded to be::

[[2, 1, 0, 0], [3, 2, 2, 0], [1, 2, 1, 3]]

When blank_label is $"last"$, we can index the labels as {'a': 0, 'b': 1, 'c': 2}, and we reserve the channel index 3 for blank label in data tensor. The resulting label tensor should be padded to be::

[[1, 0, -1, -1], [2, 1, 1, -1], [0, 1, 0, 2]]

$out$ is a list of CTC loss values, one per example in the batch.

See Connectionist Temporal Classification: Labelling Unsegmented Sequence Data with Recurrent Neural Networks, A. Graves et al. for more information on the definition and the algorithm.

Defined in src/operator/contrib/ctc_loss.cc:L115

Arguments

data::NDArray-or-SymbolicNode: Input data to the ctc_loss op.
label::NDArray-or-SymbolicNode: Ground-truth labels for the loss.
data_lengths::NDArray-or-SymbolicNode: Lengths of data for each of the samples. Only required when use_data_lengths is true.
label_lengths::NDArray-or-SymbolicNode: Lengths of labels for each of the samples. Only required when use_label_lengths is true.
use_data_lengths::boolean, optional, default=0: Whether the data lenghts are decided by data_lengths. If false, the lengths are equal to the max sequence length.
use_label_lengths::boolean, optional, default=0: Whether the label lenghts are decided by label_lengths, or derived from padding_mask. If false, the lengths are derived from the first occurrence of the value of padding_mask. The value of padding_mask is $0$ when first CTC label is reserved for blank, and $-1$ when last label is reserved for blank. See blank_label.
blank_label::{'first', 'last'},optional, default='first': Set the label that is reserved for blank label.If "first", 0-th label is reserved, and label values for tokens in the vocabulary are between $1$ and $alphabet_size-1$, and the padding mask is $-1$. If "last", last label value $alphabet_size-1$ is reserved for blank label instead, and label values for tokens in the vocabulary are between $0$ and $alphabet_size-2$, and the padding mask is $0$.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._contrib_dequantize — Method.

_contrib_dequantize(input, min_range, max_range, out_type)

Dequantize the input tensor into a float tensor. [min_range, max_range] are scalar floats that spcify the range for the output data.

Each value of the tensor will undergo the following:

out[i] = min_range + (in[i] * (max_range - min_range) / range(INPUT_TYPE))

here range(T) = numeric_limits<T>::max() - numeric_limits<T>::min()

Defined in src/operator/contrib/dequantize.cc:L41

Arguments

input::NDArray-or-SymbolicNode: A ndarray/symbol of type uint8
min_range::NDArray-or-SymbolicNode: The minimum scalar value possibly produced for the input
max_range::NDArray-or-SymbolicNode: The maximum scalar value possibly produced for the input
out_type::{'float32'}, required: Output data type.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._contrib_fft — Method.

_contrib_fft(data, compute_size)

Apply 1D FFT to input"

.. note:: fft is only available on GPU.

Currently accept 2 input data shapes: (N, d) or (N1, N2, N3, d), data can only be real numbers. The output data has shape: (N, 2d) or (N1, N2, N3, 2d). The format is: [real0, imag0, real1, imag1, ...].

Example::

data = np.random.normal(0,1,(3,4)) out = mx.contrib.ndarray.fft(data = mx.nd.array(data,ctx = mx.gpu(0)))

Defined in src/operator/contrib/fft.cc:L56

Arguments

data::NDArray-or-SymbolicNode: Input data to the FFTOp.
compute_size::int, optional, default='128': Maximum size of sub-batch to be forwarded at one time
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._contrib_ifft — Method.

_contrib_ifft(data, compute_size)

Apply 1D ifft to input"

.. note:: ifft is only available on GPU.

Currently accept 2 input data shapes: (N, d) or (N1, N2, N3, d). Data is in format: [real0, imag0, real1, imag1, ...]. Last dimension must be an even number. The output data has shape: (N, d/2) or (N1, N2, N3, d/2). It is only the real part of the result.

Example::

data = np.random.normal(0,1,(3,4)) out = mx.contrib.ndarray.ifft(data = mx.nd.array(data,ctx = mx.gpu(0)))

Defined in src/operator/contrib/ifft.cc:L58

Arguments

data::NDArray-or-SymbolicNode: Input data to the IFFTOp.
compute_size::int, optional, default='128': Maximum size of sub-batch to be forwarded at one time
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._contrib_quantize — Method.

_contrib_quantize(input, min_range, max_range, out_type)

Quantize a input tensor from float to out_type, with user-specified min_range and max_range.

[min_range, max_range] are scalar floats that spcify the range for the input data. Each value of the tensor will undergo the following:

out[i] = (in[i] - min_range) * range(OUTPUT_TYPE) / (max_range - min_range)

here range(T) = numeric_limits<T>::max() - numeric_limits<T>::min()

Defined in src/operator/contrib/quantize.cc:L41

Arguments

input::NDArray-or-SymbolicNode: A ndarray/symbol of type float32
min_range::NDArray-or-SymbolicNode: The minimum scalar value possibly produced for the input
max_range::NDArray-or-SymbolicNode: The maximum scalar value possibly produced for the input
out_type::{'uint8'},optional, default='uint8': Output data type.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._copy — Method.

_copy(data)

Returns a copy of the input.

From:src/operator/tensor/elemwise_unary_op_basic.cc:111

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._copyto — Method.

_copyto(data)

Arguments

data::NDArray: input data
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._crop_assign — Method.

_crop_assign(lhs, rhs, begin, end, step)

_crop_assign is an alias of _slice_assign.

Assign the rhs to a cropped subset of lhs.

Requirements

output should be explicitly given and be the same as lhs.
lhs and rhs are of the same data type, and on the same device.

From:src/operator/tensor/matrix_op.cc:385

Arguments

lhs::NDArray-or-SymbolicNode: Source input
rhs::NDArray-or-SymbolicNode: value to assign
begin::Shape(tuple), required: starting indices for the slice operation, supports negative indices.
end::Shape(tuple), required: ending indices for the slice operation, supports negative indices.
step::Shape(tuple), optional, default=[]: step for the slice operation, supports negative values.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._crop_assign_scalar — Method.

_crop_assign_scalar(data, scalar, begin, end, step)

_crop_assign_scalar is an alias of _slice_assign_scalar.

(Assign the scalar to a cropped subset of the input.

Requirements

output should be explicitly given and be the same as input

)

From:src/operator/tensor/matrix_op.cc:410

Arguments

data::NDArray-or-SymbolicNode: Source input
scalar::float, optional, default=0: The scalar value for assignment.
begin::Shape(tuple), required: starting indices for the slice operation, supports negative indices.
end::Shape(tuple), required: ending indices for the slice operation, supports negative indices.
step::Shape(tuple), optional, default=[]: step for the slice operation, supports negative values.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._cvcopyMakeBorder — Method.

_cvcopyMakeBorder(src, top, bot, left, right, type, value, values)

Pad image border with OpenCV.

Arguments

src::NDArray: source image
top::int, required: Top margin.
bot::int, required: Bottom margin.
left::int, required: Left margin.
right::int, required: Right margin.
type::int, optional, default='0': Filling type (default=cv2.BORDER_CONSTANT).
value::double, optional, default=0: (Deprecated! Use $values$ instead.) Fill with single value.
values::tuple of <double>, optional, default=[]: Fill with value(RGB[A] or gray), up to 4 channels.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._cvimdecode — Method.

_cvimdecode(buf, flag, to_rgb)

Decode image with OpenCV. Note: return image in RGB by default, instead of OpenCV's default BGR.

Arguments

buf::NDArray: Buffer containing binary encoded image
flag::int, optional, default='1': Convert decoded image to grayscale (0) or color (1).
to_rgb::boolean, optional, default=1: Whether to convert decoded image to mxnet's default RGB format (instead of opencv's default BGR).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._cvimread — Method.

_cvimread(filename, flag, to_rgb)

Read and decode image with OpenCV. Note: return image in RGB by default, instead of OpenCV's default BGR.

Arguments

filename::string, required: Name of the image file to be loaded.
flag::int, optional, default='1': Convert decoded image to grayscale (0) or color (1).
to_rgb::boolean, optional, default=1: Whether to convert decoded image to mxnet's default RGB format (instead of opencv's default BGR).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._cvimresize — Method.

_cvimresize(src, w, h, interp)

Resize image with OpenCV.

Arguments

src::NDArray: source image
w::int, required: Width of resized image.
h::int, required: Height of resized image.
interp::int, optional, default='1': Interpolation method (default=cv2.INTER_LINEAR).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._div_scalar — Method.

_div_scalar(data, scalar)

Divide an array with a scalar.

$_div_scalar$ only operates on data array of input if input is sparse.

For example, if input of shape (100, 100) has only 2 non zero elements, i.e. input.data = [5, 6], scalar = nan, it will result output.data = [nan, nan] instead of 10000 nans.

Defined in src/operator/tensor/elemwise_binary_scalar_op_basic.cc:L171

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._equal — Method.

_equal(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._equal_scalar — Method.

_equal_scalar(data, scalar)

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._full — Method.

_full(shape, ctx, dtype, value)

fill target with a scalar value

Arguments

shape::Shape(tuple), optional, default=[]: The shape of the output
ctx::string, optional, default='': Context of output, in format cpu|gpu|cpu_pinned.Only used for imperative calls.
dtype::{'float16', 'float32', 'float64', 'int32', 'uint8'},optional, default='float32': Target data type.
value::double, required: Value with which to fill newly created tensor
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._grad_add — Method.

_grad_add(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._greater — Method.

_greater(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._greater_equal — Method.

_greater_equal(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._greater_equal_scalar — Method.

_greater_equal_scalar(data, scalar)

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._greater_scalar — Method.

_greater_scalar(data, scalar)

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._hypot — Method.

_hypot(lhs, rhs)

Given the "legs" of a right triangle, return its hypotenuse.

Defined in src/operator/tensor/elemwise_binary_op_extended.cc:L79

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._hypot_scalar — Method.

_hypot_scalar(data, scalar)

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._identity_with_attr_like_rhs — Method.

_identity_with_attr_like_rhs(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: First input.
rhs::NDArray-or-SymbolicNode: Second input.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._image_adjust_lighting — Method.

_image_adjust_lighting(data, alpha)

Adjust the lighting level of the input. Follow the AlexNet style.

Defined in src/operator/image/image_random.cc:L118

Arguments

data::NDArray-or-SymbolicNode: The input.
alpha::tuple of <float>, required: The lighting alphas for the R, G, B channels.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._image_flip_left_right — Method.

_image_flip_left_right(data)

Defined in src/operator/image/image_random.cc:L68

Arguments

data::NDArray-or-SymbolicNode: The input.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._image_flip_top_bottom — Method.

_image_flip_top_bottom(data)

Defined in src/operator/image/image_random.cc:L76

Arguments

data::NDArray-or-SymbolicNode: The input.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._image_normalize — Method.

_image_normalize(data, mean, std)

Defined in src/operator/image/image_random.cc:L52

Arguments

data::NDArray-or-SymbolicNode: The input.
mean::tuple of <float>, required: Sequence of mean for each channel.
std::tuple of <float>, required: Sequence of standard deviations for each channel.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._image_random_brightness — Method.

_image_random_brightness(data, min_factor, max_factor)

Defined in src/operator/image/image_random.cc:L84

Arguments

data::NDArray-or-SymbolicNode: The input.
min_factor::float, required: Minimum factor.
max_factor::float, required: Maximum factor.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._image_random_color_jitter — Method.

_image_random_color_jitter(data, brightness, contrast, saturation, hue)

Defined in src/operator/image/image_random.cc:L111

Arguments

data::NDArray-or-SymbolicNode: The input.
brightness::float, required: How much to jitter brightness.
contrast::float, required: How much to jitter contrast.
saturation::float, required: How much to jitter saturation.
hue::float, required: How much to jitter hue.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._image_random_contrast — Method.

_image_random_contrast(data, min_factor, max_factor)

Defined in src/operator/image/image_random.cc:L90

Arguments

data::NDArray-or-SymbolicNode: The input.
min_factor::float, required: Minimum factor.
max_factor::float, required: Maximum factor.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._image_random_flip_left_right — Method.

_image_random_flip_left_right(data)

Defined in src/operator/image/image_random.cc:L72

Arguments

data::NDArray-or-SymbolicNode: The input.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._image_random_flip_top_bottom — Method.

_image_random_flip_top_bottom(data)

Defined in src/operator/image/image_random.cc:L80

Arguments

data::NDArray-or-SymbolicNode: The input.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._image_random_hue — Method.

_image_random_hue(data, min_factor, max_factor)

Defined in src/operator/image/image_random.cc:L104

Arguments

data::NDArray-or-SymbolicNode: The input.
min_factor::float, required: Minimum factor.
max_factor::float, required: Maximum factor.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._image_random_lighting — Method.

_image_random_lighting(data, alpha_std)

Randomly add PCA noise. Follow the AlexNet style.

Defined in src/operator/image/image_random.cc:L125

Arguments

data::NDArray-or-SymbolicNode: The input.
alpha_std::float, optional, default=0.05: Level of the lighting noise.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._image_random_saturation — Method.

_image_random_saturation(data, min_factor, max_factor)

Defined in src/operator/image/image_random.cc:L97

Arguments

data::NDArray-or-SymbolicNode: The input.
min_factor::float, required: Minimum factor.
max_factor::float, required: Maximum factor.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._image_to_tensor — Method.

_image_to_tensor(data)

Defined in src/operator/image/image_random.cc:L42

Arguments

data::NDArray-or-SymbolicNode: The input.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._imdecode — Method.

_imdecode(mean, index, x0, y0, x1, y1, c, size)

Decode an image, clip to (x0, y0, x1, y1), subtract mean, and write to buffer

Arguments

mean::NDArray-or-SymbolicNode: image mean
index::int: buffer position for output
x0::int: x0
y0::int: y0
x1::int: x1
y1::int: y1
c::int: channel
size::int: length of str_img
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._lesser — Method.

_lesser(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._lesser_equal — Method.

_lesser_equal(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._lesser_equal_scalar — Method.

_lesser_equal_scalar(data, scalar)

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._lesser_scalar — Method.

_lesser_scalar(data, scalar)

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._linalg_gelqf — Method.

_linalg_gelqf(A)

LQ factorization for general matrix. Input is a tensor A of dimension n >= 2.

If n=2, we compute the LQ factorization (LAPACK gelqf, followed by orglq). A must have shape (x, y) with x <= y, and must have full rank =x. The LQ factorization consists of L with shape (x, x) and Q with shape (x, y), so that:

A = L * Q

Here, L is lower triangular (upper triangle equal to zero) with nonzero diagonal, and Q is row-orthonormal, meaning that

Q * Q\ :sup:T

is equal to the identity matrix of shape (x, x).

If n>2, gelqf is performed separately on the trailing two dimensions for all inputs (batch mode).

.. note:: The operator supports float32 and float64 data types only.

Examples::

// Single LQ factorization A = [[1., 2., 3.], [4., 5., 6.]] Q, L = gelqf(A) Q = [[-0.26726124, -0.53452248, -0.80178373], [0.87287156, 0.21821789, -0.43643578]] L = [[-3.74165739, 0.], [-8.55235974, 1.96396101]]

// Batch LQ factorization A = [[[1., 2., 3.], [4., 5., 6.]], [[7., 8., 9.], [10., 11., 12.]]] Q, L = gelqf(A) Q = [[[-0.26726124, -0.53452248, -0.80178373], [0.87287156, 0.21821789, -0.43643578]], [[-0.50257071, -0.57436653, -0.64616234], [0.7620735, 0.05862104, -0.64483142]]] L = [[[-3.74165739, 0.], [-8.55235974, 1.96396101]], [[-13.92838828, 0.], [-19.09768702, 0.52758934]]]

Defined in src/operator/tensor/la_op.cc:L529

Arguments

A::NDArray-or-SymbolicNode: Tensor of input matrices to be factorized
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._linalg_gemm — Method.

_linalg_gemm(A, B, C, transpose_a, transpose_b, alpha, beta)

Performs general matrix multiplication and accumulation. Input are tensors A, B, C, each of dimension n >= 2 and having the same shape on the leading n-2 dimensions.

If n=2, the BLAS3 function gemm is performed:

out = alpha * op\ (A) * op\ (B) + beta * C

Here, alpha and beta are scalar parameters, and op() is either the identity or matrix transposition (depending on transpose_a, transpose_b).

If n>2, gemm is performed separately on the trailing two dimensions for all inputs (batch mode).

.. note:: The operator supports float32 and float64 data types only.

Examples::

// Single matrix multiply-add A = [[1.0, 1.0], [1.0, 1.0]] B = [[1.0, 1.0], [1.0, 1.0], [1.0, 1.0]] C = [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0]] gemm(A, B, C, transpose_b=True, alpha=2.0, beta=10.0) = [[14.0, 14.0, 14.0], [14.0, 14.0, 14.0]]

// Batch matrix multiply-add A = [[[1.0, 1.0]], [[0.1, 0.1]]] B = [[[1.0, 1.0]], [[0.1, 0.1]]] C = [[[10.0]], [[0.01]]] gemm(A, B, C, transpose_b=True, alpha=2.0 , beta=10.0) = [[[104.0]], [[0.14]]]

Defined in src/operator/tensor/la_op.cc:L69

Arguments

A::NDArray-or-SymbolicNode: Tensor of input matrices
B::NDArray-or-SymbolicNode: Tensor of input matrices
C::NDArray-or-SymbolicNode: Tensor of input matrices
transpose_a::boolean, optional, default=0: Multiply with transposed of first input (A).
transpose_b::boolean, optional, default=0: Multiply with transposed of second input (B).
alpha::double, optional, default=1: Scalar factor multiplied with A*B.
beta::double, optional, default=1: Scalar factor multiplied with C.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._linalg_gemm2 — Method.

_linalg_gemm2(A, B, transpose_a, transpose_b, alpha)

Performs general matrix multiplication. Input are tensors A, B, each of dimension n >= 2 and having the same shape on the leading n-2 dimensions.

If n=2, the BLAS3 function gemm is performed:

out = alpha * op\ (A) * op\ (B)

Here alpha is a scalar parameter and op() is either the identity or the matrix transposition (depending on transpose_a, transpose_b).

If n>2, gemm is performed separately on the trailing two dimensions for all inputs (batch mode).

.. note:: The operator supports float32 and float64 data types only.

Examples::

// Single matrix multiply A = [[1.0, 1.0], [1.0, 1.0]] B = [[1.0, 1.0], [1.0, 1.0], [1.0, 1.0]] gemm2(A, B, transpose_b=True, alpha=2.0) = [[4.0, 4.0, 4.0], [4.0, 4.0, 4.0]]

// Batch matrix multiply A = [[[1.0, 1.0]], [[0.1, 0.1]]] B = [[[1.0, 1.0]], [[0.1, 0.1]]] gemm2(A, B, transpose_b=True, alpha=2.0) = [[[4.0]], [[0.04 ]]]

Defined in src/operator/tensor/la_op.cc:L128

Arguments

A::NDArray-or-SymbolicNode: Tensor of input matrices
B::NDArray-or-SymbolicNode: Tensor of input matrices
transpose_a::boolean, optional, default=0: Multiply with transposed of first input (A).
transpose_b::boolean, optional, default=0: Multiply with transposed of second input (B).
alpha::double, optional, default=1: Scalar factor multiplied with A*B.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._linalg_potrf — Method.

_linalg_potrf(A)

Performs Cholesky factorization of a symmetric positive-definite matrix. Input is a tensor A of dimension n >= 2.

If n=2, the Cholesky factor L of the symmetric, positive definite matrix A is computed. L is lower triangular (entries of upper triangle are all zero), has positive diagonal entries, and:

A = L * L\ :sup:T

If n>2, potrf is performed separately on the trailing two dimensions for all inputs (batch mode).

.. note:: The operator supports float32 and float64 data types only.

Examples::

// Single matrix factorization A = [[4.0, 1.0], [1.0, 4.25]] potrf(A) = [[2.0, 0], [0.5, 2.0]]

// Batch matrix factorization A = [[[4.0, 1.0], [1.0, 4.25]], [[16.0, 4.0], [4.0, 17.0]]] potrf(A) = [[[2.0, 0], [0.5, 2.0]], [[4.0, 0], [1.0, 4.0]]]

Defined in src/operator/tensor/la_op.cc:L178

Arguments

A::NDArray-or-SymbolicNode: Tensor of input matrices to be decomposed
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._linalg_potri — Method.

_linalg_potri(A)

Performs matrix inversion from a Cholesky factorization. Input is a tensor A of dimension n >= 2.

If n=2, A is a lower triangular matrix (entries of upper triangle are all zero) with positive diagonal. We compute:

out = A\ :sup:-T * A\ :sup:-1

In other words, if A is the Cholesky factor of a symmetric positive definite matrix B (obtained by potrf), then

out = B\ :sup:-1

If n>2, potri is performed separately on the trailing two dimensions for all inputs (batch mode).

.. note:: The operator supports float32 and float64 data types only.

.. note:: Use this operator only if you are certain you need the inverse of B, and cannot use the Cholesky factor A (potrf), together with backsubstitution (trsm). The latter is numerically much safer, and also cheaper.

Examples::

// Single matrix inverse A = [[2.0, 0], [0.5, 2.0]] potri(A) = [[0.26563, -0.0625], [-0.0625, 0.25]]

// Batch matrix inverse A = [[[2.0, 0], [0.5, 2.0]], [[4.0, 0], [1.0, 4.0]]] potri(A) = [[[0.26563, -0.0625], [-0.0625, 0.25]], [[0.06641, -0.01562], [-0.01562, 0,0625]]]

Defined in src/operator/tensor/la_op.cc:L236

Arguments

A::NDArray-or-SymbolicNode: Tensor of lower triangular matrices
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._linalg_sumlogdiag — Method.

_linalg_sumlogdiag(A)

Computes the sum of the logarithms of the diagonal elements of a square matrix. Input is a tensor A of dimension n >= 2.

If n=2, A must be square with positive diagonal entries. We sum the natural logarithms of the diagonal elements, the result has shape (1,).

If n>2, sumlogdiag is performed separately on the trailing two dimensions for all inputs (batch mode).

.. note:: The operator supports float32 and float64 data types only.

Examples::

// Single matrix reduction A = [[1.0, 1.0], [1.0, 7.0]] sumlogdiag(A) = [1.9459]

// Batch matrix reduction A = [[[1.0, 1.0], [1.0, 7.0]], [[3.0, 0], [0, 17.0]]] sumlogdiag(A) = [1.9459, 3.9318]

Defined in src/operator/tensor/la_op.cc:L405

Arguments

A::NDArray-or-SymbolicNode: Tensor of square matrices
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._linalg_syevd — Method.

_linalg_syevd(A)

Eigendecomposition for symmetric matrix. Input is a tensor A of dimension n >= 2.

If n=2, A must be symmetric, of shape (x, x). We compute the eigendecomposition, resulting in the orthonormal matrix U of eigenvectors, shape (x, x), and the vector L of eigenvalues, shape (x,), so that:

U * A = diag(L) * U

Here:

U * U\ :sup:T = U\ :sup:T * U = I

where I is the identity matrix. Also, L(0) <= L(1) <= L(2) <= ... (ascending order).

If n>2, syevd is performed separately on the trailing two dimensions of A (batch mode). In this case, U has n dimensions like A, and L has n-1 dimensions.

.. note:: The operator supports float32 and float64 data types only.

.. note:: Derivatives for this operator are defined only if A is such that all its eigenvalues are distinct, and the eigengaps are not too small. If you need gradients, do not apply this operator to matrices with multiple eigenvalues.

Examples::

// Single symmetric eigendecomposition A = [[1., 2.], [2., 4.]] U, L = syevd(A) U = [[0.89442719, -0.4472136], [0.4472136, 0.89442719]] L = [0., 5.]

// Batch symmetric eigendecomposition A = [[[1., 2.], [2., 4.]], [[1., 2.], [2., 5.]]] U, L = syevd(A) U = [[[0.89442719, -0.4472136], [0.4472136, 0.89442719]], [[0.92387953, -0.38268343], [0.38268343, 0.92387953]]] L = [[0., 5.], [0.17157288, 5.82842712]]

Defined in src/operator/tensor/la_op.cc:L598

Arguments

A::NDArray-or-SymbolicNode: Tensor of input matrices to be factorized
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._linalg_syrk — Method.

_linalg_syrk(A, transpose, alpha)

Multiplication of matrix with its transpose. Input is a tensor A of dimension n >= 2.

If n=2, the operator performs the BLAS3 function syrk:

out = alpha * A * A\ :sup:T

if transpose=False, or

out = alpha * A\ :sup:T \ * A

if transpose=True.

If n>2, syrk is performed separately on the trailing two dimensions for all inputs (batch mode).

.. note:: The operator supports float32 and float64 data types only.

Examples::

// Single matrix multiply A = [[1., 2., 3.], [4., 5., 6.]] syrk(A, alpha=1., transpose=False) = [[14., 32.], [32., 77.]] syrk(A, alpha=1., transpose=True) = [[17., 22., 27.], [22., 29., 36.], [27., 36., 45.]]

// Batch matrix multiply A = [[[1., 1.]], [[0.1, 0.1]]] syrk(A, alpha=2., transpose=False) = [[[4.]], [[0.04]]]

Defined in src/operator/tensor/la_op.cc:L461

Arguments

A::NDArray-or-SymbolicNode: Tensor of input matrices
transpose::boolean, optional, default=0: Use transpose of input matrix.
alpha::double, optional, default=1: Scalar factor to be applied to the result.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._linalg_trmm — Method.

_linalg_trmm(A, B, transpose, rightside, alpha)

Performs multiplication with a lower triangular matrix. Input are tensors A, B, each of dimension n >= 2 and having the same shape on the leading n-2 dimensions.

If n=2, A must be lower triangular. The operator performs the BLAS3 function trmm:

out = alpha * op\ (A) * B

if rightside=False, or

out = alpha * B * op\ (A)

if rightside=True. Here, alpha is a scalar parameter, and op() is either the identity or the matrix transposition (depending on transpose).

If n>2, trmm is performed separately on the trailing two dimensions for all inputs (batch mode).

.. note:: The operator supports float32 and float64 data types only.

Examples::

// Single triangular matrix multiply A = [[1.0, 0], [1.0, 1.0]] B = [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0]] trmm(A, B, alpha=2.0) = [[2.0, 2.0, 2.0], [4.0, 4.0, 4.0]]

// Batch triangular matrix multiply A = [[[1.0, 0], [1.0, 1.0]], [[1.0, 0], [1.0, 1.0]]] B = [[[1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], [[0.5, 0.5, 0.5], [0.5, 0.5, 0.5]]] trmm(A, B, alpha=2.0) = [[[2.0, 2.0, 2.0], [4.0, 4.0, 4.0]], [[1.0, 1.0, 1.0], [2.0, 2.0, 2.0]]]

Defined in src/operator/tensor/la_op.cc:L293

Arguments

A::NDArray-or-SymbolicNode: Tensor of lower triangular matrices
B::NDArray-or-SymbolicNode: Tensor of matrices
transpose::boolean, optional, default=0: Use transposed of the triangular matrix
rightside::boolean, optional, default=0: Multiply triangular matrix from the right to non-triangular one.
alpha::double, optional, default=1: Scalar factor to be applied to the result.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._linalg_trsm — Method.

_linalg_trsm(A, B, transpose, rightside, alpha)

Solves matrix equation involving a lower triangular matrix. Input are tensors A, B, each of dimension n >= 2 and having the same shape on the leading n-2 dimensions.

If n=2, A must be lower triangular. The operator performs the BLAS3 function trsm, solving for out in:

op\ (A) * out = alpha * B

if rightside=False, or

out * op\ (A) = alpha * B

if rightside=True. Here, alpha is a scalar parameter, and op() is either the identity or the matrix transposition (depending on transpose).

If n>2, trsm is performed separately on the trailing two dimensions for all inputs (batch mode).

.. note:: The operator supports float32 and float64 data types only.

Examples::

// Single matrix solve A = [[1.0, 0], [1.0, 1.0]] B = [[2.0, 2.0, 2.0], [4.0, 4.0, 4.0]] trsm(A, B, alpha=0.5) = [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0]]

// Batch matrix solve A = [[[1.0, 0], [1.0, 1.0]], [[1.0, 0], [1.0, 1.0]]] B = [[[2.0, 2.0, 2.0], [4.0, 4.0, 4.0]], [[4.0, 4.0, 4.0], [8.0, 8.0, 8.0]]] trsm(A, B, alpha=0.5) = [[[1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], [[2.0, 2.0, 2.0], [2.0, 2.0, 2.0]]]

Defined in src/operator/tensor/la_op.cc:L356

Arguments

A::NDArray-or-SymbolicNode: Tensor of lower triangular matrices
B::NDArray-or-SymbolicNode: Tensor of matrices
transpose::boolean, optional, default=0: Use transposed of the triangular matrix
rightside::boolean, optional, default=0: Multiply triangular matrix from the right to non-triangular one.
alpha::double, optional, default=1: Scalar factor to be applied to the result.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._maximum — Method.

_maximum(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._maximum_scalar — Method.

_maximum_scalar(data, scalar)

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._minimum — Method.

_minimum(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._minimum_scalar — Method.

_minimum_scalar(data, scalar)

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._minus — Method.

_minus(lhs, rhs)

_minus is an alias of elemwise_sub.

Subtracts arguments element-wise.

The storage type of $elemwise_sub$ output depends on storage types of inputs

elemwise_sub(row_sparse, row_sparse) = row_sparse
elemwise_sub(csr, csr) = csr
otherwise, $elemwise_sub$ generates output with default storage

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._minus_scalar — Method.

_minus_scalar(data, scalar)

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._mod — Method.

_mod(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._mod_scalar — Method.

_mod_scalar(data, scalar)

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._mul — Method.

_mul(lhs, rhs)

_mul is an alias of elemwise_mul.

Multiplies arguments element-wise.

The storage type of $elemwise_mul$ output depends on storage types of inputs

elemwise_mul(default, default) = default
elemwise_mul(row_sparse, row_sparse) = row_sparse
elemwise_mul(default, row_sparse) = default
elemwise_mul(row_sparse, default) = default
elemwise_mul(csr, csr) = csr
otherwise, $elemwise_mul$ generates output with default storage

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._mul_scalar — Method.

_mul_scalar(data, scalar)

Multiply an array with a scalar.

$_mul_scalar$ only operates on data array of input if input is sparse.

For example, if input of shape (100, 100) has only 2 non zero elements, i.e. input.data = [5, 6], scalar = nan, it will result output.data = [nan, nan] instead of 10000 nans.

Defined in src/operator/tensor/elemwise_binary_scalar_op_basic.cc:L149

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._not_equal — Method.

_not_equal(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._not_equal_scalar — Method.

_not_equal_scalar(data, scalar)

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._onehot_encode — Method.

_onehot_encode(lhs, rhs)

Arguments

lhs::NDArray: Left operand to the function.
rhs::NDArray: Right operand to the function.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._ones — Method.

_ones(shape, ctx, dtype)

fill target with ones

Arguments

shape::Shape(tuple), optional, default=[]: The shape of the output
ctx::string, optional, default='': Context of output, in format cpu|gpu|cpu_pinned.Only used for imperative calls.
dtype::{'float16', 'float32', 'float64', 'int32', 'uint8'},optional, default='float32': Target data type.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._plus — Method.

_plus(lhs, rhs)

_plus is an alias of elemwise_add.

Adds arguments element-wise.

The storage type of $elemwise_add$ output depends on storage types of inputs

elemwise_add(row_sparse, row_sparse) = row_sparse
elemwise_add(csr, csr) = csr
otherwise, $elemwise_add$ generates output with default storage

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._plus_scalar — Method.

_plus_scalar(data, scalar)

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._power — Method.

_power(lhs, rhs)

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._power_scalar — Method.

_power_scalar(data, scalar)

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._random_exponential — Method.

_random_exponential(lam, shape, ctx, dtype)

Draw random samples from an exponential distribution.

Samples are distributed according to an exponential distribution parametrized by lambda (rate).

Example::

exponential(lam=4, shape=(2,2)) = [[ 0.0097189 , 0.08999364], [ 0.04146638, 0.31715935]]

Defined in src/operator/random/sample_op.cc:L115

Arguments

lam::float, optional, default=1: Lambda parameter (rate) of the exponential distribution.
shape::Shape(tuple), optional, default=[]: Shape of the output.
ctx::string, optional, default='': Context of output, in format cpu|gpu|cpu_pinned. Only used for imperative calls.
dtype::{'None', 'float16', 'float32', 'float64'},optional, default='None': DType of the output in case this can't be inferred. Defaults to float32 if not defined (dtype=None).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._random_gamma — Method.

_random_gamma(alpha, beta, shape, ctx, dtype)

Draw random samples from a gamma distribution.

Samples are distributed according to a gamma distribution parametrized by alpha (shape) and beta (scale).

Example::

gamma(alpha=9, beta=0.5, shape=(2,2)) = [[ 7.10486984, 3.37695289], [ 3.91697288, 3.65933681]]

Defined in src/operator/random/sample_op.cc:L100

Arguments

alpha::float, optional, default=1: Alpha parameter (shape) of the gamma distribution.
beta::float, optional, default=1: Beta parameter (scale) of the gamma distribution.
shape::Shape(tuple), optional, default=[]: Shape of the output.
ctx::string, optional, default='': Context of output, in format cpu|gpu|cpu_pinned. Only used for imperative calls.
dtype::{'None', 'float16', 'float32', 'float64'},optional, default='None': DType of the output in case this can't be inferred. Defaults to float32 if not defined (dtype=None).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._random_generalized_negative_binomial — Method.

_random_generalized_negative_binomial(mu, alpha, shape, ctx, dtype)

Draw random samples from a generalized negative binomial distribution.

Samples are distributed according to a generalized negative binomial distribution parametrized by mu (mean) and alpha (dispersion). alpha is defined as 1/k where k is the failure limit of the number of unsuccessful experiments (generalized to real numbers). Samples will always be returned as a floating point data type.

Example::

generalized_negative_binomial(mu=2.0, alpha=0.3, shape=(2,2)) = [[ 2., 1.], [ 6., 4.]]

Defined in src/operator/random/sample_op.cc:L168

Arguments

mu::float, optional, default=1: Mean of the negative binomial distribution.
alpha::float, optional, default=1: Alpha (dispersion) parameter of the negative binomial distribution.
shape::Shape(tuple), optional, default=[]: Shape of the output.
ctx::string, optional, default='': Context of output, in format cpu|gpu|cpu_pinned. Only used for imperative calls.
dtype::{'None', 'float16', 'float32', 'float64'},optional, default='None': DType of the output in case this can't be inferred. Defaults to float32 if not defined (dtype=None).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._random_negative_binomial — Method.

_random_negative_binomial(k, p, shape, ctx, dtype)

Draw random samples from a negative binomial distribution.

Samples are distributed according to a negative binomial distribution parametrized by k (limit of unsuccessful experiments) and p (failure probability in each experiment). Samples will always be returned as a floating point data type.

Example::

negative_binomial(k=3, p=0.4, shape=(2,2)) = [[ 4., 7.], [ 2., 5.]]

Defined in src/operator/random/sample_op.cc:L149

Arguments

k::int, optional, default='1': Limit of unsuccessful experiments.
p::float, optional, default=1: Failure probability in each experiment.
shape::Shape(tuple), optional, default=[]: Shape of the output.
ctx::string, optional, default='': Context of output, in format cpu|gpu|cpu_pinned. Only used for imperative calls.
dtype::{'None', 'float16', 'float32', 'float64'},optional, default='None': DType of the output in case this can't be inferred. Defaults to float32 if not defined (dtype=None).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._random_normal — Method.

_random_normal(loc, scale, shape, ctx, dtype)

Draw random samples from a normal (Gaussian) distribution.

.. note:: The existing alias $normal$ is deprecated.

Samples are distributed according to a normal distribution parametrized by loc (mean) and scale (standard deviation).

Example::

normal(loc=0, scale=1, shape=(2,2)) = [[ 1.89171135, -1.16881478], [-1.23474145, 1.55807114]]

Defined in src/operator/random/sample_op.cc:L85

Arguments

loc::float, optional, default=0: Mean of the distribution.
scale::float, optional, default=1: Standard deviation of the distribution.
shape::Shape(tuple), optional, default=[]: Shape of the output.
ctx::string, optional, default='': Context of output, in format cpu|gpu|cpu_pinned. Only used for imperative calls.
dtype::{'None', 'float16', 'float32', 'float64'},optional, default='None': DType of the output in case this can't be inferred. Defaults to float32 if not defined (dtype=None).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._random_poisson — Method.

_random_poisson(lam, shape, ctx, dtype)

Draw random samples from a Poisson distribution.

Samples are distributed according to a Poisson distribution parametrized by lambda (rate). Samples will always be returned as a floating point data type.

Example::

poisson(lam=4, shape=(2,2)) = [[ 5., 2.], [ 4., 6.]]

Defined in src/operator/random/sample_op.cc:L132

Arguments

lam::float, optional, default=1: Lambda parameter (rate) of the Poisson distribution.
shape::Shape(tuple), optional, default=[]: Shape of the output.
ctx::string, optional, default='': Context of output, in format cpu|gpu|cpu_pinned. Only used for imperative calls.
dtype::{'None', 'float16', 'float32', 'float64'},optional, default='None': DType of the output in case this can't be inferred. Defaults to float32 if not defined (dtype=None).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._random_uniform — Method.

_random_uniform(low, high, shape, ctx, dtype)

Draw random samples from a uniform distribution.

.. note:: The existing alias $uniform$ is deprecated.

Samples are uniformly distributed over the half-open interval [low, high) (includes low, but excludes high).

Example::

uniform(low=0, high=1, shape=(2,2)) = [[ 0.60276335, 0.85794562], [ 0.54488319, 0.84725171]]

Defined in src/operator/random/sample_op.cc:L66

Arguments

low::float, optional, default=0: Lower bound of the distribution.
high::float, optional, default=1: Upper bound of the distribution.
shape::Shape(tuple), optional, default=[]: Shape of the output.
ctx::string, optional, default='': Context of output, in format cpu|gpu|cpu_pinned. Only used for imperative calls.
dtype::{'None', 'float16', 'float32', 'float64'},optional, default='None': DType of the output in case this can't be inferred. Defaults to float32 if not defined (dtype=None).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._rdiv_scalar — Method.

_rdiv_scalar(data, scalar)

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._rminus_scalar — Method.

_rminus_scalar(data, scalar)

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._rmod_scalar — Method.

_rmod_scalar(data, scalar)

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._rpower_scalar — Method.

_rpower_scalar(data, scalar)

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sample_exponential — Method.

_sample_exponential(lam, shape, dtype)

Concurrent sampling from multiple exponential distributions with parameters lambda (rate).

The parameters of the distributions are provided as an input array. Let [s] be the shape of the input array, n be the dimension of [s], [t] be the shape specified as the parameter of the operator, and m be the dimension of [t]. Then the output will be a (n+m)-dimensional array with shape [s]x[t].

For any valid n-dimensional index i with respect to the input array, output[i] will be an m-dimensional array that holds randomly drawn samples from the distribution which is parameterized by the input value at index i. If the shape parameter of the operator is not set, then one sample will be drawn per distribution and the output array has the same shape as the input array.

Examples::

lam = [ 1.0, 8.5 ]

// Draw a single sample for each distribution sample_exponential(lam) = [ 0.51837951, 0.09994757]

// Draw a vector containing two samples for each distribution sample_exponential(lam, shape=(2)) = [[ 0.51837951, 0.19866663], [ 0.09994757, 0.50447971]]

Defined in src/operator/random/multisample_op.cc:L284

Arguments

lam::NDArray-or-SymbolicNode: Lambda (rate) parameters of the distributions.
shape::Shape(tuple), optional, default=[]: Shape to be sampled from each random distribution.
dtype::{'None', 'float16', 'float32', 'float64'},optional, default='None': DType of the output in case this can't be inferred. Defaults to float32 if not defined (dtype=None).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sample_gamma — Method.

_sample_gamma(alpha, shape, dtype, beta)

Concurrent sampling from multiple gamma distributions with parameters alpha (shape) and beta (scale).

The parameters of the distributions are provided as input arrays. Let [s] be the shape of the input arrays, n be the dimension of [s], [t] be the shape specified as the parameter of the operator, and m be the dimension of [t]. Then the output will be a (n+m)-dimensional array with shape [s]x[t].

For any valid n-dimensional index i with respect to the input arrays, output[i] will be an m-dimensional array that holds randomly drawn samples from the distribution which is parameterized by the input values at index i. If the shape parameter of the operator is not set, then one sample will be drawn per distribution and the output array has the same shape as the input arrays.

Examples::

alpha = [ 0.0, 2.5 ] beta = [ 1.0, 0.7 ]

// Draw a single sample for each distribution sample_gamma(alpha, beta) = [ 0. , 2.25797319]

// Draw a vector containing two samples for each distribution sample_gamma(alpha, beta, shape=(2)) = [[ 0. , 0. ], [ 2.25797319, 1.70734084]]

Defined in src/operator/random/multisample_op.cc:L282

Arguments

alpha::NDArray-or-SymbolicNode: Alpha (shape) parameters of the distributions.
shape::Shape(tuple), optional, default=[]: Shape to be sampled from each random distribution.
dtype::{'None', 'float16', 'float32', 'float64'},optional, default='None': DType of the output in case this can't be inferred. Defaults to float32 if not defined (dtype=None).
beta::NDArray-or-SymbolicNode: Beta (scale) parameters of the distributions.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sample_generalized_negative_binomial — Method.

_sample_generalized_negative_binomial(mu, shape, dtype, alpha)

Concurrent sampling from multiple generalized negative binomial distributions with parameters mu (mean) and alpha (dispersion).

The parameters of the distributions are provided as input arrays. Let [s] be the shape of the input arrays, n be the dimension of [s], [t] be the shape specified as the parameter of the operator, and m be the dimension of [t]. Then the output will be a (n+m)-dimensional array with shape [s]x[t].

For any valid n-dimensional index i with respect to the input arrays, output[i] will be an m-dimensional array that holds randomly drawn samples from the distribution which is parameterized by the input values at index i. If the shape parameter of the operator is not set, then one sample will be drawn per distribution and the output array has the same shape as the input arrays.

Samples will always be returned as a floating point data type.

Examples::

mu = [ 2.0, 2.5 ] alpha = [ 1.0, 0.1 ]

// Draw a single sample for each distribution sample_generalized_negative_binomial(mu, alpha) = [ 0., 3.]

// Draw a vector containing two samples for each distribution sample_generalized_negative_binomial(mu, alpha, shape=(2)) = [[ 0., 3.], [ 3., 1.]]

Defined in src/operator/random/multisample_op.cc:L293

Arguments

mu::NDArray-or-SymbolicNode: Means of the distributions.
shape::Shape(tuple), optional, default=[]: Shape to be sampled from each random distribution.
dtype::{'None', 'float16', 'float32', 'float64'},optional, default='None': DType of the output in case this can't be inferred. Defaults to float32 if not defined (dtype=None).
alpha::NDArray-or-SymbolicNode: Alpha (dispersion) parameters of the distributions.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sample_multinomial — Method.

_sample_multinomial(data, shape, get_prob, dtype)

Concurrent sampling from multiple multinomial distributions.

data is an n dimensional array whose last dimension has length k, where k is the number of possible outcomes of each multinomial distribution. This operator will draw shape samples from each distribution. If shape is empty one sample will be drawn from each distribution.

If get_prob is true, a second array containing log likelihood of the drawn samples will also be returned. This is usually used for reinforcement learning where you can provide reward as head gradient for this array to estimate gradient.

Note that the input distribution must be normalized, i.e. data must sum to 1 along its last axis.

Examples::

probs = [[0, 0.1, 0.2, 0.3, 0.4], [0.4, 0.3, 0.2, 0.1, 0]]

// Draw a single sample for each distribution sample_multinomial(probs) = [3, 0]

// Draw a vector containing two samples for each distribution sample_multinomial(probs, shape=(2)) = [[4, 2], [0, 0]]

// requests log likelihood sample_multinomial(probs, get_prob=True) = [2, 1], [0.2, 0.3]

Arguments

data::NDArray-or-SymbolicNode: Distribution probabilities. Must sum to one on the last axis.
shape::Shape(tuple), optional, default=[]: Shape to be sampled from each random distribution.
get_prob::boolean, optional, default=0: Whether to also return the log probability of sampled result. This is usually used for differentiating through stochastic variables, e.g. in reinforcement learning.
dtype::{'int32'},optional, default='int32': DType of the output in case this can't be inferred. Only support int32 for now.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sample_negative_binomial — Method.

_sample_negative_binomial(k, shape, dtype, p)

Concurrent sampling from multiple negative binomial distributions with parameters k (failure limit) and p (failure probability).

The parameters of the distributions are provided as input arrays. Let [s] be the shape of the input arrays, n be the dimension of [s], [t] be the shape specified as the parameter of the operator, and m be the dimension of [t]. Then the output will be a (n+m)-dimensional array with shape [s]x[t].

For any valid n-dimensional index i with respect to the input arrays, output[i] will be an m-dimensional array that holds randomly drawn samples from the distribution which is parameterized by the input values at index i. If the shape parameter of the operator is not set, then one sample will be drawn per distribution and the output array has the same shape as the input arrays.

Samples will always be returned as a floating point data type.

Examples::

k = [ 20, 49 ] p = [ 0.4 , 0.77 ]

// Draw a single sample for each distribution sample_negative_binomial(k, p) = [ 15., 16.]

// Draw a vector containing two samples for each distribution sample_negative_binomial(k, p, shape=(2)) = [[ 15., 50.], [ 16., 12.]]

Defined in src/operator/random/multisample_op.cc:L289

Arguments

k::NDArray-or-SymbolicNode: Limits of unsuccessful experiments.
shape::Shape(tuple), optional, default=[]: Shape to be sampled from each random distribution.
dtype::{'None', 'float16', 'float32', 'float64'},optional, default='None': DType of the output in case this can't be inferred. Defaults to float32 if not defined (dtype=None).
p::NDArray-or-SymbolicNode: Failure probabilities in each experiment.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sample_normal — Method.

_sample_normal(mu, shape, dtype, sigma)

Concurrent sampling from multiple normal distributions with parameters mu (mean) and sigma (standard deviation).

The parameters of the distributions are provided as input arrays. Let [s] be the shape of the input arrays, n be the dimension of [s], [t] be the shape specified as the parameter of the operator, and m be the dimension of [t]. Then the output will be a (n+m)-dimensional array with shape [s]x[t].

For any valid n-dimensional index i with respect to the input arrays, output[i] will be an m-dimensional array that holds randomly drawn samples from the distribution which is parameterized by the input values at index i. If the shape parameter of the operator is not set, then one sample will be drawn per distribution and the output array has the same shape as the input arrays.

Examples::

mu = [ 0.0, 2.5 ] sigma = [ 1.0, 3.7 ]

// Draw a single sample for each distribution sample_normal(mu, sigma) = [-0.56410581, 0.95934606]

// Draw a vector containing two samples for each distribution sample_normal(mu, sigma, shape=(2)) = [[-0.56410581, 0.2928229 ], [ 0.95934606, 4.48287058]]

Defined in src/operator/random/multisample_op.cc:L279

Arguments

mu::NDArray-or-SymbolicNode: Means of the distributions.
shape::Shape(tuple), optional, default=[]: Shape to be sampled from each random distribution.
dtype::{'None', 'float16', 'float32', 'float64'},optional, default='None': DType of the output in case this can't be inferred. Defaults to float32 if not defined (dtype=None).
sigma::NDArray-or-SymbolicNode: Standard deviations of the distributions.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sample_poisson — Method.

_sample_poisson(lam, shape, dtype)

Concurrent sampling from multiple Poisson distributions with parameters lambda (rate).

The parameters of the distributions are provided as an input array. Let [s] be the shape of the input array, n be the dimension of [s], [t] be the shape specified as the parameter of the operator, and m be the dimension of [t]. Then the output will be a (n+m)-dimensional array with shape [s]x[t].

For any valid n-dimensional index i with respect to the input array, output[i] will be an m-dimensional array that holds randomly drawn samples from the distribution which is parameterized by the input value at index i. If the shape parameter of the operator is not set, then one sample will be drawn per distribution and the output array has the same shape as the input array.

Samples will always be returned as a floating point data type.

Examples::

lam = [ 1.0, 8.5 ]

// Draw a single sample for each distribution sample_poisson(lam) = [ 0., 13.]

// Draw a vector containing two samples for each distribution sample_poisson(lam, shape=(2)) = [[ 0., 4.], [ 13., 8.]]

Defined in src/operator/random/multisample_op.cc:L286

Arguments

lam::NDArray-or-SymbolicNode: Lambda (rate) parameters of the distributions.
shape::Shape(tuple), optional, default=[]: Shape to be sampled from each random distribution.
dtype::{'None', 'float16', 'float32', 'float64'},optional, default='None': DType of the output in case this can't be inferred. Defaults to float32 if not defined (dtype=None).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sample_uniform — Method.

_sample_uniform(low, shape, dtype, high)

Concurrent sampling from multiple uniform distributions on the intervals given by [low,high).

The parameters of the distributions are provided as input arrays. Let [s] be the shape of the input arrays, n be the dimension of [s], [t] be the shape specified as the parameter of the operator, and m be the dimension of [t]. Then the output will be a (n+m)-dimensional array with shape [s]x[t].

For any valid n-dimensional index i with respect to the input arrays, output[i] will be an m-dimensional array that holds randomly drawn samples from the distribution which is parameterized by the input values at index i. If the shape parameter of the operator is not set, then one sample will be drawn per distribution and the output array has the same shape as the input arrays.

Examples::

low = [ 0.0, 2.5 ] high = [ 1.0, 3.7 ]

// Draw a single sample for each distribution sample_uniform(low, high) = [ 0.40451524, 3.18687344]

// Draw a vector containing two samples for each distribution sample_uniform(low, high, shape=(2)) = [[ 0.40451524, 0.18017688], [ 3.18687344, 3.68352246]]

Defined in src/operator/random/multisample_op.cc:L277

Arguments

low::NDArray-or-SymbolicNode: Lower bounds of the distributions.
shape::Shape(tuple), optional, default=[]: Shape to be sampled from each random distribution.
dtype::{'None', 'float16', 'float32', 'float64'},optional, default='None': DType of the output in case this can't be inferred. Defaults to float32 if not defined (dtype=None).
high::NDArray-or-SymbolicNode: Upper bounds of the distributions.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._scatter_elemwise_div — Method.

_scatter_elemwise_div(lhs, rhs)

Divides arguments element-wise. If the left-hand-side input is 'row_sparse', then only the values which exist in the left-hand sparse array are computed. The 'missing' values are ignored.

The storage type of $_scatter_elemwise_div$ output depends on storage types of inputs

_scatter_elemwise_div(row_sparse, row_sparse) = row_sparse
_scatter_elemwise_div(row_sparse, dense) = row_sparse
_scatter_elemwise_div(row_sparse, csr) = row_sparse
otherwise, $_scatter_elemwise_div$ behaves exactly like elemwise_div and generates output

with default storage

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._scatter_minus_scalar — Method.

_scatter_minus_scalar(data, scalar)

Subtracts a scalar to a tensor element-wise. If the left-hand-side input is 'row_sparse' or 'csr', then only the values which exist in the left-hand sparse array are computed. The 'missing' values are ignored.

The storage type of $_scatter_minus_scalar$ output depends on storage types of inputs

_scatter_minus_scalar(row_sparse, scalar) = row_sparse
_scatter_minus_scalar(csr, scalar) = csr
otherwise, $_scatter_minus_scalar$ behaves exactly like _minus_scalar and generates output

with default storage

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._scatter_plus_scalar — Method.

_scatter_plus_scalar(data, scalar)

Adds a scalar to a tensor element-wise. If the left-hand-side input is 'row_sparse' or 'csr', then only the values which exist in the left-hand sparse array are computed. The 'missing' values are ignored.

The storage type of $_scatter_plus_scalar$ output depends on storage types of inputs

_scatter_plus_scalar(row_sparse, scalar) = row_sparse
_scatter_plus_scalar(csr, scalar) = csr
otherwise, $_scatter_plus_scalar$ behaves exactly like _plus_scalar and generates output

with default storage

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._scatter_set_nd — Method.

_scatter_set_nd(data, indices, shape)

This operator has the same functionality as scatter_nd except that it does not reset the elements not indexed by the input index NDArray in the input data NDArray. .. note:: This operator is for internal use only.

Examples::

data = [2, 3, 0] indices = [[1, 1, 0], [0, 1, 0]] out = [[1, 1], [1, 1]] scatter_nd(data=data, indices=indices, out=out) out = [[0, 1], [2, 3]]

Arguments

data::NDArray-or-SymbolicNode: data
indices::NDArray-or-SymbolicNode: indices
shape::Shape(tuple), required: Shape of output.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._slice_assign — Method.

_slice_assign(lhs, rhs, begin, end, step)

Assign the rhs to a cropped subset of lhs.

Requirements

output should be explicitly given and be the same as lhs.
lhs and rhs are of the same data type, and on the same device.

From:src/operator/tensor/matrix_op.cc:385

Arguments

lhs::NDArray-or-SymbolicNode: Source input
rhs::NDArray-or-SymbolicNode: value to assign
begin::Shape(tuple), required: starting indices for the slice operation, supports negative indices.
end::Shape(tuple), required: ending indices for the slice operation, supports negative indices.
step::Shape(tuple), optional, default=[]: step for the slice operation, supports negative values.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._slice_assign_scalar — Method.

_slice_assign_scalar(data, scalar, begin, end, step)

(Assign the scalar to a cropped subset of the input.

Requirements

output should be explicitly given and be the same as input

)

From:src/operator/tensor/matrix_op.cc:410

Arguments

data::NDArray-or-SymbolicNode: Source input
scalar::float, optional, default=0: The scalar value for assignment.
begin::Shape(tuple), required: starting indices for the slice operation, supports negative indices.
end::Shape(tuple), required: ending indices for the slice operation, supports negative indices.
step::Shape(tuple), optional, default=[]: step for the slice operation, supports negative values.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_ElementWiseSum — Method.

_sparse_ElementWiseSum(args)

_sparse_ElementWiseSum is an alias of add_n.

Note: _sparse_ElementWiseSum takes variable number of positional inputs. So instead of calling as _sparse_ElementWiseSum([x, y, z], num_args=3), one should call via _sparse_ElementWiseSum(x, y, z), and num_args will be determined automatically.

Adds all input arguments element-wise.

.. math:: add_n(a_1, a_2, ..., a_n) = a_1 + a_2 + ... + a_n

$add_n$ is potentially more efficient than calling $add$ by n times.

The storage type of $add_n$ output depends on storage types of inputs

add_n(row_sparse, row_sparse, ..) = row_sparse
otherwise, $add_n$ generates output with default storage

Defined in src/operator/tensor/elemwise_sum.cc:L123

Arguments

args::NDArray-or-SymbolicNode[]: Positional input arguments
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_abs — Method.

_sparse_abs(data)

_sparse_abs is an alias of abs.

Returns element-wise absolute value of the input.

Example::

abs([-2, 0, 3]) = [2, 0, 3]

The storage type of $abs$ output depends upon the input storage type:

abs(default) = default
abs(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L385

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_adam_update — Method.

_sparse_adam_update(weight, grad, mean, var, lr, beta1, beta2, epsilon, wd, rescale_grad, clip_gradient)

_sparse_adam_update is an alias of adam_update.

Update function for Adam optimizer. Adam is seen as a generalization of AdaGrad.

Adam update consists of the following steps, where g represents gradient and m, v are 1st and 2nd order moment estimates (mean and variance).

.. math::

g_t = \nabla J(W_{t-1})\ m_t = \beta_1 m_{t-1} + (1 - \beta_1) g_t\ v_t = \beta_2 v_{t-1} + (1 - \beta_2) g_t^2\ W_t = W_{t-1} - \alpha \frac{ m_t }{ \sqrt{ v_t } + \epsilon }

It updates the weights using::

m = beta1m + (1-beta1)grad v = beta2v + (1-beta2)(grad**2) w += - learning_rate * m / (sqrt(v) + epsilon)

If w, m and v are all of $row_sparse$ storage type, only the row slices whose indices appear in grad.indices are updated (for w, m and v)::

for row in grad.indices: m[row] = beta1m[row] + (1-beta1)grad[row] v[row] = beta2v[row] + (1-beta2)(grad[row]**2) w[row] += - learning_rate * m[row] / (sqrt(v[row]) + epsilon)

Defined in src/operator/optimizer_op.cc:L383

Arguments

weight::NDArray-or-SymbolicNode: Weight
grad::NDArray-or-SymbolicNode: Gradient
mean::NDArray-or-SymbolicNode: Moving mean
var::NDArray-or-SymbolicNode: Moving variance
lr::float, required: Learning rate
beta1::float, optional, default=0.9: The decay rate for the 1st moment estimates.
beta2::float, optional, default=0.999: The decay rate for the 2nd moment estimates.
epsilon::float, optional, default=1e-08: A small constant for numerical stability.
wd::float, optional, default=0: Weight decay augments the objective function with a regularization term that penalizes large weights. The penalty scales with the square of the magnitude of each weight.
rescale_grad::float, optional, default=1: Rescale gradient to grad = rescale_grad*grad.
clip_gradient::float, optional, default=-1: Clip gradient to the range of [-clip_gradient, clip_gradient] If clip_gradient <= 0, gradient clipping is turned off. grad = max(min(grad, clip_gradient), -clip_gradient).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_add_n — Method.

_sparse_add_n(args)

_sparse_add_n is an alias of add_n.

Note: _sparse_add_n takes variable number of positional inputs. So instead of calling as _sparse_add_n([x, y, z], num_args=3), one should call via _sparse_add_n(x, y, z), and num_args will be determined automatically.

Adds all input arguments element-wise.

.. math:: add_n(a_1, a_2, ..., a_n) = a_1 + a_2 + ... + a_n

$add_n$ is potentially more efficient than calling $add$ by n times.

The storage type of $add_n$ output depends on storage types of inputs

add_n(row_sparse, row_sparse, ..) = row_sparse
otherwise, $add_n$ generates output with default storage

Defined in src/operator/tensor/elemwise_sum.cc:L123

Arguments

args::NDArray-or-SymbolicNode[]: Positional input arguments
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_arccos — Method.

_sparse_arccos(data)

_sparse_arccos is an alias of arccos.

Returns element-wise inverse cosine of the input array.

The input should be in range [-1, 1]. The output is in the closed interval :math:[0, \pi]

.. math:: arccos([-1, -.707, 0, .707, 1]) = [\pi, 3\pi/4, \pi/2, \pi/4, 0]

The storage type of $arccos$ output is always dense

Defined in src/operator/tensor/elemwise_unary_op_trig.cc:L123

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_arccosh — Method.

_sparse_arccosh(data)

_sparse_arccosh is an alias of arccosh.

Returns the element-wise inverse hyperbolic cosine of the input array, computed element-wise.

The storage type of $arccosh$ output is always dense

Defined in src/operator/tensor/elemwise_unary_op_trig.cc:L264

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_arcsin — Method.

_sparse_arcsin(data)

_sparse_arcsin is an alias of arcsin.

Returns element-wise inverse sine of the input array.

The input should be in the range [-1, 1]. The output is in the closed interval of [:math:-\pi/2, :math:\pi/2].

.. math:: arcsin([-1, -.707, 0, .707, 1]) = [-\pi/2, -\pi/4, 0, \pi/4, \pi/2]

The storage type of $arcsin$ output depends upon the input storage type:

arcsin(default) = default
arcsin(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_trig.cc:L104

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_arcsinh — Method.

_sparse_arcsinh(data)

_sparse_arcsinh is an alias of arcsinh.

Returns the element-wise inverse hyperbolic sine of the input array, computed element-wise.

The storage type of $arcsinh$ output depends upon the input storage type:

arcsinh(default) = default
arcsinh(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_trig.cc:L250

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_arctan — Method.

_sparse_arctan(data)

_sparse_arctan is an alias of arctan.

Returns element-wise inverse tangent of the input array.

The output is in the closed interval :math:[-\pi/2, \pi/2]

.. math:: arctan([-1, 0, 1]) = [-\pi/4, 0, \pi/4]

The storage type of $arctan$ output depends upon the input storage type:

arctan(default) = default
arctan(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_trig.cc:L144

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_arctanh — Method.

_sparse_arctanh(data)

_sparse_arctanh is an alias of arctanh.

Returns the element-wise inverse hyperbolic tangent of the input array, computed element-wise.

The storage type of $arctanh$ output depends upon the input storage type:

arctanh(default) = default
arctanh(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_trig.cc:L281

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_cast_storage — Method.

_sparse_cast_storage(data, stype)

_sparse_cast_storage is an alias of cast_storage.

Casts tensor storage type to the new type.

When an NDArray with default storage type is cast to csr or row_sparse storage, the result is compact, which means:

for csr, zero values will not be retained
for row_sparse, row slices of all zeros will not be retained

The storage type of $cast_storage$ output depends on stype parameter:

cast_storage(csr, 'default') = default
cast_storage(row_sparse, 'default') = default
cast_storage(default, 'csr') = csr
cast_storage(default, 'row_sparse') = row_sparse

Example::

dense = [[ 0.,  1.,  0.],
         [ 2.,  0.,  3.],
         [ 0.,  0.,  0.],
         [ 0.,  0.,  0.]]

# cast to row_sparse storage type
rsp = cast_storage(dense, 'row_sparse')
rsp.indices = [0, 1]
rsp.values = [[ 0.,  1.,  0.],
              [ 2.,  0.,  3.]]

# cast to csr storage type
csr = cast_storage(dense, 'csr')
csr.indices = [1, 0, 2]
csr.values = [ 1.,  2.,  3.]
csr.indptr = [0, 1, 3, 3, 3]

Defined in src/operator/tensor/cast_storage.cc:L69

Arguments

data::NDArray-or-SymbolicNode: The input.
stype::{'csr', 'default', 'row_sparse'}, required: Output storage type.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_ceil — Method.

_sparse_ceil(data)

_sparse_ceil is an alias of ceil.

Returns element-wise ceiling of the input.

The ceil of the scalar x is the smallest integer i, such that i >= x.

Example::

ceil([-2.1, -1.9, 1.5, 1.9, 2.1]) = [-2., -1., 2., 2., 3.]

The storage type of $ceil$ output depends upon the input storage type:

ceil(default) = default
ceil(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L463

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_clip — Method.

_sparse_clip(data, a_min, a_max)

_sparse_clip is an alias of clip.

Clips (limits) the values in an array.

Given an interval, values outside the interval are clipped to the interval edges. Clipping $x$ between a_min and a_x would be::

clip(x, a_min, a_max) = max(min(x, a_max), a_min))

Example::

x = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]

clip(x,1,8) = [ 1.,  1.,  2.,  3.,  4.,  5.,  6.,  7.,  8.,  8.]

The storage type of $clip$ output depends on storage types of inputs and the a_min, a_max parameter values:

clip(default) = default
clip(row_sparse, a_min <= 0, a_max >= 0) = row_sparse
clip(csr, a_min <= 0, a_max >= 0) = csr
clip(row_sparse, a_min < 0, a_max < 0) = default
clip(row_sparse, a_min > 0, a_max > 0) = default
clip(csr, a_min < 0, a_max < 0) = csr
clip(csr, a_min > 0, a_max > 0) = csr

Defined in src/operator/tensor/matrix_op.cc:L490

Arguments

data::NDArray-or-SymbolicNode: Input array.
a_min::float, required: Minimum value
a_max::float, required: Maximum value
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_cos — Method.

_sparse_cos(data)

_sparse_cos is an alias of cos.

Computes the element-wise cosine of the input array.

The input should be in radians (:math:2\pi rad equals 360 degrees).

.. math:: cos([0, \pi/4, \pi/2]) = [1, 0.707, 0]

The storage type of $cos$ output is always dense

Defined in src/operator/tensor/elemwise_unary_op_trig.cc:L63

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_cosh — Method.

_sparse_cosh(data)

_sparse_cosh is an alias of cosh.

Returns the hyperbolic cosine of the input array, computed element-wise.

.. math:: cosh(x) = 0.5\times(exp(x) + exp(-x))

The storage type of $cosh$ output is always dense

Defined in src/operator/tensor/elemwise_unary_op_trig.cc:L216

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_degrees — Method.

_sparse_degrees(data)

_sparse_degrees is an alias of degrees.

Converts each element of the input array from radians to degrees.

.. math:: degrees([0, \pi/2, \pi, 3\pi/2, 2\pi]) = [0, 90, 180, 270, 360]

The storage type of $degrees$ output depends upon the input storage type:

degrees(default) = default
degrees(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_trig.cc:L163

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_dot — Method.

_sparse_dot(lhs, rhs, transpose_a, transpose_b)

_sparse_dot is an alias of dot.

Dot product of two arrays.

$dot$'s behavior depends on the input array dimensions:

1-D arrays: inner product of vectors
2-D arrays: matrix multiplication
N-D arrays: a sum product over the last axis of the first input and the first axis of the second input

For example, given 3-D $x$ with shape (n,m,k) and $y$ with shape (k,r,s), the result array will have shape (n,m,r,s). It is computed by::

dot(x,y)[i,j,a,b] = sum(x[i,j,:]*y[:,a,b])

Example::

x = reshape([0,1,2,3,4,5,6,7], shape=(2,2,2)) y = reshape([7,6,5,4,3,2,1,0], shape=(2,2,2)) dot(x,y)[0,0,1,1] = 0 sum(x[0,0,:]*y[:,1,1]) = 0

The storage type of $dot$ output depends on storage types of inputs and transpose options:

dot(csr, default) = default
dot(csr.T, default) = row_sparse
dot(csr, row_sparse) = default
dot(default, csr) = csr
otherwise, $dot$ generates output with default storage

Defined in src/operator/tensor/dot.cc:L62

Arguments

lhs::NDArray-or-SymbolicNode: The first input
rhs::NDArray-or-SymbolicNode: The second input
transpose_a::boolean, optional, default=0: If true then transpose the first input before dot.
transpose_b::boolean, optional, default=0: If true then transpose the second input before dot.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_elemwise_add — Method.

_sparse_elemwise_add(lhs, rhs)

_sparse_elemwise_add is an alias of elemwise_add.

Adds arguments element-wise.

The storage type of $elemwise_add$ output depends on storage types of inputs

elemwise_add(row_sparse, row_sparse) = row_sparse
elemwise_add(csr, csr) = csr
otherwise, $elemwise_add$ generates output with default storage

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_elemwise_div — Method.

_sparse_elemwise_div(lhs, rhs)

_sparse_elemwise_div is an alias of elemwise_div.

Divides arguments element-wise.

The storage type of $elemwise_div$ output is always dense

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_elemwise_mul — Method.

_sparse_elemwise_mul(lhs, rhs)

_sparse_elemwise_mul is an alias of elemwise_mul.

Multiplies arguments element-wise.

The storage type of $elemwise_mul$ output depends on storage types of inputs

elemwise_mul(default, default) = default
elemwise_mul(row_sparse, row_sparse) = row_sparse
elemwise_mul(default, row_sparse) = default
elemwise_mul(row_sparse, default) = default
elemwise_mul(csr, csr) = csr
otherwise, $elemwise_mul$ generates output with default storage

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_elemwise_sub — Method.

_sparse_elemwise_sub(lhs, rhs)

_sparse_elemwise_sub is an alias of elemwise_sub.

Subtracts arguments element-wise.

The storage type of $elemwise_sub$ output depends on storage types of inputs

elemwise_sub(row_sparse, row_sparse) = row_sparse
elemwise_sub(csr, csr) = csr
otherwise, $elemwise_sub$ generates output with default storage

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_exp — Method.

_sparse_exp(data)

_sparse_exp is an alias of exp.

Returns element-wise exponential value of the input.

.. math:: exp(x) = e^x \approx 2.718^x

Example::

exp([0, 1, 2]) = [1., 2.71828175, 7.38905621]

The storage type of $exp$ output is always dense

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L641

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_expm1 — Method.

_sparse_expm1(data)

_sparse_expm1 is an alias of expm1.

Returns $exp(x) - 1$ computed element-wise on the input.

This function provides greater precision than $exp(x) - 1$ for small values of $x$.

The storage type of $expm1$ output depends upon the input storage type:

expm1(default) = default
expm1(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L720

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_fix — Method.

_sparse_fix(data)

_sparse_fix is an alias of fix.

Returns element-wise rounded value to the nearest integer towards zero of the input.

Example::

fix([-2.1, -1.9, 1.9, 2.1]) = [-2., -1., 1., 2.]

The storage type of $fix$ output depends upon the input storage type:

fix(default) = default
fix(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L520

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_floor — Method.

_sparse_floor(data)

_sparse_floor is an alias of floor.

Returns element-wise floor of the input.

The floor of the scalar x is the largest integer i, such that i <= x.

Example::

floor([-2.1, -1.9, 1.5, 1.9, 2.1]) = [-3., -2., 1., 1., 2.]

The storage type of $floor$ output depends upon the input storage type:

floor(default) = default
floor(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L482

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_ftrl_update — Method.

_sparse_ftrl_update(weight, grad, z, n, lr, lamda1, beta, wd, rescale_grad, clip_gradient)

_sparse_ftrl_update is an alias of ftrl_update.

Update function for Ftrl optimizer. Referenced from Ad Click Prediction: a View from the Trenches, available at http://dl.acm.org/citation.cfm?id=2488200.

It updates the weights using::

rescaled_grad = clip(grad * rescale_grad, clip_gradient) z += rescaled_grad - (sqrt(n + rescaled_grad2) - sqrt(n)) * weight / learning_rate n += rescaled_grad2 w = (sign(z) * lamda1 - z) / ((beta + sqrt(n)) / learning_rate + wd) * (abs(z) > lamda1)

If w, z and n are all of $row_sparse$ storage type, only the row slices whose indices appear in grad.indices are updated (for w, z and n)::

for row in grad.indices: rescaled_grad[row] = clip(grad[row] * rescale_grad, clip_gradient) z[row] += rescaled_grad[row] - (sqrt(n[row] + rescaled_grad[row]2) - sqrt(n[row])) * weight[row] / learning_rate n[row] += rescaled_grad[row]2 w[row] = (sign(z[row]) * lamda1 - z[row]) / ((beta + sqrt(n[row])) / learning_rate + wd) * (abs(z[row]) > lamda1)

Defined in src/operator/optimizer_op.cc:L520

Arguments

weight::NDArray-or-SymbolicNode: Weight
grad::NDArray-or-SymbolicNode: Gradient
z::NDArray-or-SymbolicNode: z
n::NDArray-or-SymbolicNode: Square of grad
lr::float, required: Learning rate
lamda1::float, optional, default=0.01: The L1 regularization coefficient.
beta::float, optional, default=1: Per-Coordinate Learning Rate beta.
wd::float, optional, default=0: Weight decay augments the objective function with a regularization term that penalizes large weights. The penalty scales with the square of the magnitude of each weight.
rescale_grad::float, optional, default=1: Rescale gradient to grad = rescale_grad*grad.
clip_gradient::float, optional, default=-1: Clip gradient to the range of [-clip_gradient, clip_gradient] If clip_gradient <= 0, gradient clipping is turned off. grad = max(min(grad, clip_gradient), -clip_gradient).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_gamma — Method.

_sparse_gamma(data)

_sparse_gamma is an alias of gamma.

Returns the gamma function (extension of the factorial function to the reals), computed element-wise on the input array.

The storage type of $gamma$ output is always dense

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_gammaln — Method.

_sparse_gammaln(data)

_sparse_gammaln is an alias of gammaln.

Returns element-wise log of the absolute value of the gamma function of the input.

The storage type of $gammaln$ output is always dense

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_log — Method.

_sparse_log(data)

_sparse_log is an alias of log.

Returns element-wise Natural logarithmic value of the input.

The natural logarithm is logarithm in base e, so that $log(exp(x)) = x$

The storage type of $log$ output is always dense

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L653

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_log10 — Method.

_sparse_log10(data)

_sparse_log10 is an alias of log10.

Returns element-wise Base-10 logarithmic value of the input.

$10**log10(x) = x$

The storage type of $log10$ output is always dense

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L665

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_log1p — Method.

_sparse_log1p(data)

_sparse_log1p is an alias of log1p.

Returns element-wise $log(1 + x)$ value of the input.

This function is more accurate than $log(1 + x)$ for small $x$ so that :math:1+x\approx 1

The storage type of $log1p$ output depends upon the input storage type:

log1p(default) = default
log1p(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L702

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_log2 — Method.

_sparse_log2(data)

_sparse_log2 is an alias of log2.

Returns element-wise Base-2 logarithmic value of the input.

$2**log2(x) = x$

The storage type of $log2$ output is always dense

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L677

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_make_loss — Method.

_sparse_make_loss(data)

_sparse_make_loss is an alias of make_loss.

Make your own loss function in network construction.

This operator accepts a customized loss function symbol as a terminal loss and the symbol should be an operator with no backward dependency. The output of this function is the gradient of loss with respect to the input data.

For example, if you are a making a cross entropy loss function. Assume $out$ is the predicted output and $label$ is the true label, then the cross entropy can be defined as::

cross_entropy = label * log(out) + (1 - label) * log(1 - out) loss = make_loss(cross_entropy)

We will need to use $make_loss$ when we are creating our own loss function or we want to combine multiple loss functions. Also we may want to stop some variables' gradients from backpropagation. See more detail in $BlockGrad$ or $stop_gradient$.

The storage type of $make_loss$ output depends upon the input storage type:

make_loss(default) = default
make_loss(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L199

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_mean — Method.

_sparse_mean(data, axis, keepdims, exclude)

_sparse_mean is an alias of mean.

Computes the mean of array elements over given axes.

Defined in src/operator/tensor/broadcast_reduce_op_value.cc:L101

Arguments

data::NDArray-or-SymbolicNode: The input
axis::Shape(tuple), optional, default=[]: The axis or axes along which to perform the reduction.

`` The default,axis=(), will compute over all elements into a scalar array with shape(1,)`.

If axis is int, a reduction is performed on a particular axis.

If axis is a tuple of ints, a reduction is performed on all the axes specified in the tuple.

If exclude is true, reduction will be performed on the axes that are NOT in axis instead.

Negative values means indexing from right to left. `` *keepdims::boolean, optional, default=0: If this is set toTrue, the reduced axes are left in the result as dimension with size one. *exclude::boolean, optional, default=0: Whether to perform reduction on axis that are NOT in axis instead. *name::Symbol: The name of theSymbolicNode. (e.g.:my_symbol), optional. *attrs::Dict{Symbol, AbstractString}: The attributes associated with thisSymbolicNode`.

source

# MXNet.mx._sparse_negative — Method.

_sparse_negative(data)

_sparse_negative is an alias of negative.

Numerical negative of the argument, element-wise.

The storage type of $negative$ output depends upon the input storage type:

negative(default) = default
negative(row_sparse) = row_sparse
negative(csr) = csr

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_norm — Method.

_sparse_norm(data)

_sparse_norm is an alias of norm.

Flattens the input array and then computes the l2 norm.

Examples::

x = [[1, 2], [3, 4]]

norm(x) = [5.47722578]

rsp = x.cast_storage('row_sparse')

norm(rsp) = [5.47722578]

csr = x.cast_storage('csr')

norm(csr) = [5.47722578]

Defined in src/operator/tensor/broadcast_reduce_op_value.cc:L266

Arguments

data::NDArray-or-SymbolicNode: Source input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_radians — Method.

_sparse_radians(data)

_sparse_radians is an alias of radians.

Converts each element of the input array from degrees to radians.

.. math:: radians([0, 90, 180, 270, 360]) = [0, \pi/2, \pi, 3\pi/2, 2\pi]

The storage type of $radians$ output depends upon the input storage type:

radians(default) = default
radians(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_trig.cc:L182

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_relu — Method.

_sparse_relu(data)

_sparse_relu is an alias of relu.

Computes rectified linear.

.. math:: max(features, 0)

The storage type of $relu$ output depends upon the input storage type:

relu(default) = default
relu(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L83

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_retain — Method.

_sparse_retain(data, indices)

pick rows specified by user input index array from a row sparse matrix and save them in the output sparse matrix.

Example::

data = [[1, 2], [3, 4], [5, 6]] indices = [0, 1, 3] shape = (4, 2) rsp_in = row_sparse(data, indices) to_retain = [0, 3] rsp_out = retain(rsp_in, to_retain) rsp_out.values = [[1, 2], [5, 6]] rsp_out.indices = [0, 3]

The storage type of $retain$ output depends on storage types of inputs

retain(row_sparse, default) = row_sparse
otherwise, $retain$ is not supported

Defined in src/operator/tensor/sparse_retain.cc:L53

Arguments

data::NDArray-or-SymbolicNode: The input array for sparse_retain operator.
indices::NDArray-or-SymbolicNode: The index array of rows ids that will be retained.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_rint — Method.

_sparse_rint(data)

_sparse_rint is an alias of rint.

Returns element-wise rounded value to the nearest integer of the input.

.. note::

For input $n.5$ $rint$ returns $n$ while $round$ returns $n+1$.
For input $-n.5$ both $rint$ and $round$ returns $-n-1$.

Example::

rint([-1.5, 1.5, -1.9, 1.9, 2.1]) = [-2., 1., -2., 2., 2.]

The storage type of $rint$ output depends upon the input storage type:

rint(default) = default
rint(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L444

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_round — Method.

_sparse_round(data)

_sparse_round is an alias of round.

Returns element-wise rounded value to the nearest integer of the input.

Example::

round([-1.5, 1.5, -1.9, 1.9, 2.1]) = [-2., 2., -2., 2., 2.]

The storage type of $round$ output depends upon the input storage type:

round(default) = default
round(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L423

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_rsqrt — Method.

_sparse_rsqrt(data)

_sparse_rsqrt is an alias of rsqrt.

Returns element-wise inverse square-root value of the input.

.. math:: rsqrt(x) = 1/\sqrt{x}

Example::

rsqrt([4,9,16]) = [0.5, 0.33333334, 0.25]

The storage type of $rsqrt$ output is always dense

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L584

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_sgd_mom_update — Method.

_sparse_sgd_mom_update(weight, grad, mom, lr, momentum, wd, rescale_grad, clip_gradient)

_sparse_sgd_mom_update is an alias of sgd_mom_update.

Momentum update function for Stochastic Gradient Descent (SDG) optimizer.

Momentum update has better convergence rates on neural networks. Mathematically it looks like below:

.. math::

v_1 = \alpha * \nabla J(W_0)\ v_t = \gamma v_{t-1} - \alpha * \nabla J(W_{t-1})\ W_t = W_{t-1} + v_t

It updates the weights using::

v = momentum * v - learning_rate * gradient weight += v

Where the parameter $momentum$ is the decay rate of momentum estimates at each epoch.

If weight and grad are both of $row_sparse$ storage type and momentum is of $default$ storage type, standard update is applied.

If weight, grad and momentum are all of $row_sparse$ storage type, only the row slices whose indices appear in grad.indices are updated (for both weight and momentum)::

for row in gradient.indices: v[row] = momentum[row] * v[row] - learning_rate * gradient[row] weight[row] += v[row]

Defined in src/operator/optimizer_op.cc:L265

Arguments

weight::NDArray-or-SymbolicNode: Weight
grad::NDArray-or-SymbolicNode: Gradient
mom::NDArray-or-SymbolicNode: Momentum
lr::float, required: Learning rate
momentum::float, optional, default=0: The decay rate of momentum estimates at each epoch.
wd::float, optional, default=0: Weight decay augments the objective function with a regularization term that penalizes large weights. The penalty scales with the square of the magnitude of each weight.
rescale_grad::float, optional, default=1: Rescale gradient to grad = rescale_grad*grad.
clip_gradient::float, optional, default=-1: Clip gradient to the range of [-clip_gradient, clip_gradient] If clip_gradient <= 0, gradient clipping is turned off. grad = max(min(grad, clip_gradient), -clip_gradient).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_sgd_update — Method.

_sparse_sgd_update(weight, grad, lr, wd, rescale_grad, clip_gradient)

_sparse_sgd_update is an alias of sgd_update.

Update function for Stochastic Gradient Descent (SDG) optimizer.

It updates the weights using::

weight = weight - learning_rate * gradient

If weight is of $row_sparse$ storage type, only the row slices whose indices appear in grad.indices are updated::

for row in gradient.indices: weight[row] = weight[row] - learning_rate * gradient[row]

Defined in src/operator/optimizer_op.cc:L222

Arguments

weight::NDArray-or-SymbolicNode: Weight
grad::NDArray-or-SymbolicNode: Gradient
lr::float, required: Learning rate
wd::float, optional, default=0: Weight decay augments the objective function with a regularization term that penalizes large weights. The penalty scales with the square of the magnitude of each weight.
rescale_grad::float, optional, default=1: Rescale gradient to grad = rescale_grad*grad.
clip_gradient::float, optional, default=-1: Clip gradient to the range of [-clip_gradient, clip_gradient] If clip_gradient <= 0, gradient clipping is turned off. grad = max(min(grad, clip_gradient), -clip_gradient).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_sigmoid — Method.

_sparse_sigmoid(data)

_sparse_sigmoid is an alias of sigmoid.

Computes sigmoid of x element-wise.

.. math:: y = 1 / (1 + exp(-x))

The storage type of $sigmoid$ output is always dense

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L102

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_sign — Method.

_sparse_sign(data)

_sparse_sign is an alias of sign.

Returns element-wise sign of the input.

Example::

sign([-2, 0, 3]) = [-1, 0, 1]

The storage type of $sign$ output depends upon the input storage type:

sign(default) = default
sign(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L404

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_sin — Method.

_sparse_sin(data)

_sparse_sin is an alias of sin.

Computes the element-wise sine of the input array.

The input should be in radians (:math:2\pi rad equals 360 degrees).

.. math:: sin([0, \pi/4, \pi/2]) = [0, 0.707, 1]

The storage type of $sin$ output depends upon the input storage type:

sin(default) = default
sin(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_trig.cc:L46

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_sinh — Method.

_sparse_sinh(data)

_sparse_sinh is an alias of sinh.

Returns the hyperbolic sine of the input array, computed element-wise.

.. math:: sinh(x) = 0.5\times(exp(x) - exp(-x))

The storage type of $sinh$ output depends upon the input storage type:

sinh(default) = default
sinh(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_trig.cc:L201

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_slice — Method.

_sparse_slice(data, begin, end, step)

_sparse_slice is an alias of slice.

Slices a region of the array.

.. note:: $crop$ is deprecated. Use $slice$ instead.

This function returns a sliced array between the indices given by begin and end with the corresponding step.

For an input array of $shape=(d_0, d_1, ..., d_n-1)$, slice operation with $begin=(b_0, b_1...b_m-1)$, $end=(e_0, e_1, ..., e_m-1)$, and $step=(s_0, s_1, ..., s_m-1)$, where m <= n, results in an array with the shape $(|e_0-b_0|/|s_0|, ..., |e_m-1-b_m-1|/|s_m-1|, d_m, ..., d_n-1)$.

The resulting array's k-th dimension contains elements from the k-th dimension of the input array starting from index $b_k$ (inclusive) with step $s_k$ until reaching $e_k$ (exclusive).

If the k-th elements are None in the sequence of begin, end, and step, the following rule will be used to set default values. If s_k is None, set s_k=1. If s_k > 0, set b_k=0, e_k=d_k; else, set b_k=d_k-1, e_k=-1.

The storage type of $slice$ output depends on storage types of inputs

slice(csr) = csr
otherwise, $slice$ generates output with default storage

.. note:: When input data storage type is csr, it only supports step=(), or step=(None,), or step=(1,) to generate a csr output. For other step parameter values, it falls back to slicing a dense tensor.

Example::

x = [[ 1., 2., 3., 4.], [ 5., 6., 7., 8.], [ 9., 10., 11., 12.]]

slice(x, begin=(0,1), end=(2,4)) = [[ 2., 3., 4.], [ 6., 7., 8.]] slice(x, begin=(None, 0), end=(None, 3), step=(-1, 2)) = [[9., 11.], [5., 7.], [1., 3.]]

Defined in src/operator/tensor/matrix_op.cc:L359

Arguments

data::NDArray-or-SymbolicNode: Source input
begin::Shape(tuple), required: starting indices for the slice operation, supports negative indices.
end::Shape(tuple), required: ending indices for the slice operation, supports negative indices.
step::Shape(tuple), optional, default=[]: step for the slice operation, supports negative values.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_sqrt — Method.

_sparse_sqrt(data)

_sparse_sqrt is an alias of sqrt.

Returns element-wise square-root value of the input.

.. math:: \textrm{sqrt}(x) = \sqrt{x}

Example::

sqrt([4, 9, 16]) = [2, 3, 4]

The storage type of $sqrt$ output depends upon the input storage type:

sqrt(default) = default
sqrt(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L564

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_square — Method.

_sparse_square(data)

_sparse_square is an alias of square.

Returns element-wise squared value of the input.

.. math:: square(x) = x^2

Example::

square([2, 3, 4]) = [4, 9, 16]

The storage type of $square$ output depends upon the input storage type:

square(default) = default
square(row_sparse) = row_sparse
square(csr) = csr

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L541

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_stop_gradient — Method.

_sparse_stop_gradient(data)

_sparse_stop_gradient is an alias of BlockGrad.

Stops gradient computation.

Stops the accumulated gradient of the inputs from flowing through this operator in the backward direction. In other words, this operator prevents the contribution of its inputs to be taken into account for computing gradients.

Example::

v1 = [1, 2] v2 = [0, 1] a = Variable('a') b = Variable('b') b_stop_grad = stop_gradient(3 * b) loss = MakeLoss(b_stop_grad + a)

executor = loss.simple_bind(ctx=cpu(), a=(1,2), b=(1,2)) executor.forward(is_train=True, a=v1, b=v2) executor.outputs [ 1. 5.]

executor.backward() executor.grad_arrays [ 0. 0.] [ 1. 1.]

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L166

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_sum — Method.

_sparse_sum(data, axis, keepdims, exclude)

_sparse_sum is an alias of sum.

Computes the sum of array elements over given axes.

.. Note::

sum and sum_axis are equivalent. For ndarray of csr storage type summation along axis 0 and axis 1 is supported. Setting keepdims or exclude to True will cause a fallback to dense operator.

Example::

data = [[[1,2],[2,3],[1,3]], [[1,4],[4,3],[5,2]], [[7,1],[7,2],[7,3]]]

sum(data, axis=1) [[ 4. 8.] [ 10. 9.] [ 21. 6.]]

sum(data, axis=[1,2]) [ 12. 19. 27.]

data = [[1,2,0], [3,0,1], [4,1,0]]

csr = cast_storage(data, 'csr')

sum(csr, axis=0) [ 8. 3. 1.]

sum(csr, axis=1) [ 3. 4. 5.]

Defined in src/operator/tensor/broadcast_reduce_op_value.cc:L85

Arguments

data::NDArray-or-SymbolicNode: The input
axis::Shape(tuple), optional, default=[]: The axis or axes along which to perform the reduction.

`` The default,axis=(), will compute over all elements into a scalar array with shape(1,)`.

If axis is int, a reduction is performed on a particular axis.

If axis is a tuple of ints, a reduction is performed on all the axes specified in the tuple.

If exclude is true, reduction will be performed on the axes that are NOT in axis instead.

Negative values means indexing from right to left. `` *keepdims::boolean, optional, default=0: If this is set toTrue, the reduced axes are left in the result as dimension with size one. *exclude::boolean, optional, default=0: Whether to perform reduction on axis that are NOT in axis instead. *name::Symbol: The name of theSymbolicNode. (e.g.:my_symbol), optional. *attrs::Dict{Symbol, AbstractString}: The attributes associated with thisSymbolicNode`.

source

# MXNet.mx._sparse_tan — Method.

_sparse_tan(data)

_sparse_tan is an alias of tan.

Computes the element-wise tangent of the input array.

The input should be in radians (:math:2\pi rad equals 360 degrees).

.. math:: tan([0, \pi/4, \pi/2]) = [0, 1, -inf]

The storage type of $tan$ output depends upon the input storage type:

tan(default) = default
tan(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_trig.cc:L83

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_tanh — Method.

_sparse_tanh(data)

_sparse_tanh is an alias of tanh.

Returns the hyperbolic tangent of the input array, computed element-wise.

.. math:: tanh(x) = sinh(x) / cosh(x)

The storage type of $tanh$ output depends upon the input storage type:

tanh(default) = default
tanh(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_trig.cc:L234

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_trunc — Method.

_sparse_trunc(data)

_sparse_trunc is an alias of trunc.

Return the element-wise truncated value of the input.

The truncated value of the scalar x is the nearest integer i which is closer to zero than x is. In short, the fractional part of the signed number x is discarded.

Example::

trunc([-2.1, -1.9, 1.5, 1.9, 2.1]) = [-2., -1., 1., 1., 2.]

The storage type of $trunc$ output depends upon the input storage type:

trunc(default) = default
trunc(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L502

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_where — Method.

_sparse_where(condition, x, y)

_sparse_where is an alias of where.

Return the elements, either from x or y, depending on the condition.

Given three ndarrays, condition, x, and y, return an ndarray with the elements from x or y, depending on the elements from condition are true or false. x and y must have the same shape. If condition has the same shape as x, each element in the output array is from x if the corresponding element in the condition is true, and from y if false.

If condition does not have the same shape as x, it must be a 1D array whose size is the same as x's first dimension size. Each row of the output array is from x's row if the corresponding element from condition is true, and from y's row if false.

Note that all non-zero values are interpreted as $True$ in condition.

Examples::

x = [[1, 2], [3, 4]] y = [[5, 6], [7, 8]] cond = [[0, 1], [-1, 0]]

where(cond, x, y) = [[5, 2], [3, 8]]

csr_cond = cast_storage(cond, 'csr')

where(csr_cond, x, y) = [[5, 2], [3, 8]]

Defined in src/operator/tensor/control_flow_op.cc:L57

Arguments

condition::NDArray-or-SymbolicNode: condition array
x::NDArray-or-SymbolicNode:
y::NDArray-or-SymbolicNode:
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._sparse_zeros_like — Method.

_sparse_zeros_like(data)

_sparse_zeros_like is an alias of zeros_like.

Return an array of zeros with the same shape and type as the input array.

The storage type of $zeros_like$ output depends on the storage type of the input

zeros_like(row_sparse) = row_sparse
zeros_like(csr) = csr
zeros_like(default) = default

Examples::

x = [[ 1., 1., 1.], [ 1., 1., 1.]]

zeros_like(x) = [[ 0., 0., 0.], [ 0., 0., 0.]]

Arguments

data::NDArray-or-SymbolicNode: The input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx._square_sum — Method.

_square_sum(data, axis, keepdims, exclude)

Computes the square sum of array elements over a given axis for row-sparse matrix. This is a temporary solution for fusing ops square and sum together for row-sparse matrix to save memory for storing gradients. It will become deprecated once the functionality of fusing operators is finished in the future.

Example::

dns = mx.nd.array([[0, 0], [1, 2], [0, 0], [3, 4], [0, 0]]) rsp = dns.tostype('row_sparse') sum = mx.nd._internal._square_sum(rsp, axis=1) sum = [0, 5, 0, 25, 0]

Defined in src/operator/tensor/square_sum.cc:L63

Arguments

data::NDArray-or-SymbolicNode: The input
axis::Shape(tuple), optional, default=[]: The axis or axes along which to perform the reduction.

`` The default,axis=(), will compute over all elements into a scalar array with shape(1,)`.

If axis is int, a reduction is performed on a particular axis.

If axis is a tuple of ints, a reduction is performed on all the axes specified in the tuple.

If exclude is true, reduction will be performed on the axes that are NOT in axis instead.

Negative values means indexing from right to left. `` *keepdims::boolean, optional, default=0: If this is set toTrue, the reduced axes are left in the result as dimension with size one. *exclude::boolean, optional, default=0: Whether to perform reduction on axis that are NOT in axis instead. *name::Symbol: The name of theSymbolicNode. (e.g.:my_symbol), optional. *attrs::Dict{Symbol, AbstractString}: The attributes associated with thisSymbolicNode`.

source

# MXNet.mx._zeros — Method.

_zeros(shape, ctx, dtype)

fill target with zeros

Arguments

shape::Shape(tuple), optional, default=[]: The shape of the output
ctx::string, optional, default='': Context of output, in format cpu|gpu|cpu_pinned.Only used for imperative calls.
dtype::{'float16', 'float32', 'float64', 'int32', 'uint8'},optional, default='float32': Target data type.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.adam_update — Method.

adam_update(weight, grad, mean, var, lr, beta1, beta2, epsilon, wd, rescale_grad, clip_gradient)

Update function for Adam optimizer. Adam is seen as a generalization of AdaGrad.

Adam update consists of the following steps, where g represents gradient and m, v are 1st and 2nd order moment estimates (mean and variance).

.. math::

g_t = \nabla J(W_{t-1})\ m_t = \beta_1 m_{t-1} + (1 - \beta_1) g_t\ v_t = \beta_2 v_{t-1} + (1 - \beta_2) g_t^2\ W_t = W_{t-1} - \alpha \frac{ m_t }{ \sqrt{ v_t } + \epsilon }

It updates the weights using::

m = beta1m + (1-beta1)grad v = beta2v + (1-beta2)(grad**2) w += - learning_rate * m / (sqrt(v) + epsilon)

If w, m and v are all of $row_sparse$ storage type, only the row slices whose indices appear in grad.indices are updated (for w, m and v)::

for row in grad.indices: m[row] = beta1m[row] + (1-beta1)grad[row] v[row] = beta2v[row] + (1-beta2)(grad[row]**2) w[row] += - learning_rate * m[row] / (sqrt(v[row]) + epsilon)

Defined in src/operator/optimizer_op.cc:L383

Arguments

weight::NDArray-or-SymbolicNode: Weight
grad::NDArray-or-SymbolicNode: Gradient
mean::NDArray-or-SymbolicNode: Moving mean
var::NDArray-or-SymbolicNode: Moving variance
lr::float, required: Learning rate
beta1::float, optional, default=0.9: The decay rate for the 1st moment estimates.
beta2::float, optional, default=0.999: The decay rate for the 2nd moment estimates.
epsilon::float, optional, default=1e-08: A small constant for numerical stability.
wd::float, optional, default=0: Weight decay augments the objective function with a regularization term that penalizes large weights. The penalty scales with the square of the magnitude of each weight.
rescale_grad::float, optional, default=1: Rescale gradient to grad = rescale_grad*grad.
clip_gradient::float, optional, default=-1: Clip gradient to the range of [-clip_gradient, clip_gradient] If clip_gradient <= 0, gradient clipping is turned off. grad = max(min(grad, clip_gradient), -clip_gradient).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.add_n — Method.

add_n(args)

Note: add_n takes variable number of positional inputs. So instead of calling as add_n([x, y, z], num_args=3), one should call via add_n(x, y, z), and num_args will be determined automatically.

Adds all input arguments element-wise.

.. math:: add_n(a_1, a_2, ..., a_n) = a_1 + a_2 + ... + a_n

$add_n$ is potentially more efficient than calling $add$ by n times.

The storage type of $add_n$ output depends on storage types of inputs

add_n(row_sparse, row_sparse, ..) = row_sparse
otherwise, $add_n$ generates output with default storage

Defined in src/operator/tensor/elemwise_sum.cc:L123

Arguments

args::NDArray-or-SymbolicNode[]: Positional input arguments
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.argmax — Method.

argmax(data, axis, keepdims)

Returns indices of the maximum values along an axis.

In the case of multiple occurrences of maximum values, the indices corresponding to the first occurrence are returned.

Examples::

x = [[ 0., 1., 2.], [ 3., 4., 5.]]

// argmax along axis 0 argmax(x, axis=0) = [ 1., 1., 1.]

// argmax along axis 1 argmax(x, axis=1) = [ 2., 2.]

// argmax along axis 1 keeping same dims as an input array argmax(x, axis=1, keepdims=True) = [[ 2.], [ 2.]]

Defined in src/operator/tensor/broadcast_reduce_op_index.cc:L52

Arguments

data::NDArray-or-SymbolicNode: The input
axis::int or None, optional, default='None': The axis along which to perform the reduction. Negative values means indexing from right to left. $Requires axis to be set as int, because global reduction is not supported yet.$
keepdims::boolean, optional, default=0: If this is set to True, the reduced axis is left in the result as dimension with size one.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.argmax_channel — Method.

argmax_channel(data)

Returns argmax indices of each channel from the input array.

The result will be an NDArray of shape (num_channel,).

In case of multiple occurrences of the maximum values, the indices corresponding to the first occurrence are returned.

Examples::

x = [[ 0., 1., 2.], [ 3., 4., 5.]]

argmax_channel(x) = [ 2., 2.]

Defined in src/operator/tensor/broadcast_reduce_op_index.cc:L97

Arguments

data::NDArray-or-SymbolicNode: The input array
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.argmin — Method.

argmin(data, axis, keepdims)

Returns indices of the minimum values along an axis.

In the case of multiple occurrences of minimum values, the indices corresponding to the first occurrence are returned.

Examples::

x = [[ 0., 1., 2.], [ 3., 4., 5.]]

// argmin along axis 0 argmin(x, axis=0) = [ 0., 0., 0.]

// argmin along axis 1 argmin(x, axis=1) = [ 0., 0.]

// argmin along axis 1 keeping same dims as an input array argmin(x, axis=1, keepdims=True) = [[ 0.], [ 0.]]

Defined in src/operator/tensor/broadcast_reduce_op_index.cc:L77

Arguments

data::NDArray-or-SymbolicNode: The input
axis::int or None, optional, default='None': The axis along which to perform the reduction. Negative values means indexing from right to left. $Requires axis to be set as int, because global reduction is not supported yet.$
keepdims::boolean, optional, default=0: If this is set to True, the reduced axis is left in the result as dimension with size one.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.argsort — Method.

argsort(data, axis, is_ascend)

Returns the indices that would sort an input array along the given axis.

This function performs sorting along the given axis and returns an array of indices having same shape as an input array that index data in sorted order.

Examples::

x = [[ 0.3, 0.2, 0.4], [ 0.1, 0.3, 0.2]]

// sort along axis -1 argsort(x) = [[ 1., 0., 2.], [ 0., 2., 1.]]

// sort along axis 0 argsort(x, axis=0) = [[ 1., 0., 1.] [ 0., 1., 0.]]

// flatten and then sort argsort(x) = [ 3., 1., 5., 0., 4., 2.]

Defined in src/operator/tensor/ordering_op.cc:L176

Arguments

data::NDArray-or-SymbolicNode: The input array
axis::int or None, optional, default='-1': Axis along which to sort the input tensor. If not given, the flattened array is used. Default is -1.
is_ascend::boolean, optional, default=1: Whether to sort in ascending or descending order.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.batch_dot — Method.

batch_dot(lhs, rhs, transpose_a, transpose_b)

Batchwise dot product.

$batch_dot$ is used to compute dot product of $x$ and $y$ when $x$ and $y$ are data in batch, namely 3D arrays in shape of (batch_size, :, :).

For example, given $x$ with shape (batch_size, n, m) and $y$ with shape (batch_size, m, k), the result array will have shape (batch_size, n, k), which is computed by::

batch_dot(x,y)[i,:,:] = dot(x[i,:,:], y[i,:,:])

Defined in src/operator/tensor/dot.cc:L110

Arguments

lhs::NDArray-or-SymbolicNode: The first input
rhs::NDArray-or-SymbolicNode: The second input
transpose_a::boolean, optional, default=0: If true then transpose the first input before dot.
transpose_b::boolean, optional, default=0: If true then transpose the second input before dot.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.batch_take — Method.

batch_take(a, indices)

Takes elements from a data batch.

.. note:: batch_take is deprecated. Use pick instead.

Given an input array of shape $(d0, d1)$ and indices of shape $(i0,)$, the result will be an output array of shape $(i0,)$ with::

output[i] = input[i, indices[i]]

Examples::

x = [[ 1., 2.], [ 3., 4.], [ 5., 6.]]

// takes elements with specified indices batch_take(x, [0,1,0]) = [ 1. 4. 5.]

Defined in src/operator/tensor/indexing_op.cc:L426

Arguments

a::NDArray-or-SymbolicNode: The input array
indices::NDArray-or-SymbolicNode: The index array
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.broadcast_add — Method.

broadcast_add(lhs, rhs)

Returns element-wise sum of the input arrays with broadcasting.

broadcast_plus is an alias to the function broadcast_add.

Example::

x = [[ 1., 1., 1.], [ 1., 1., 1.]]

y = [[ 0.], [ 1.]]

broadcast_add(x, y) = [[ 1., 1., 1.], [ 2., 2., 2.]]

broadcast_plus(x, y) = [[ 1., 1., 1.], [ 2., 2., 2.]]

Defined in src/operator/tensor/elemwise_binary_broadcast_op_basic.cc:L51

Arguments

lhs::NDArray-or-SymbolicNode: First input to the function
rhs::NDArray-or-SymbolicNode: Second input to the function
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.broadcast_div — Method.

broadcast_div(lhs, rhs)

Returns element-wise division of the input arrays with broadcasting.

Example::

x = [[ 6., 6., 6.], [ 6., 6., 6.]]

y = [[ 2.], [ 3.]]

broadcast_div(x, y) = [[ 3., 3., 3.], [ 2., 2., 2.]]

Defined in src/operator/tensor/elemwise_binary_broadcast_op_basic.cc:L157

Arguments

lhs::NDArray-or-SymbolicNode: First input to the function
rhs::NDArray-or-SymbolicNode: Second input to the function
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.broadcast_equal — Method.

broadcast_equal(lhs, rhs)

Returns the result of element-wise equal to (==) comparison operation with broadcasting.

Example::

x = [[ 1., 1., 1.], [ 1., 1., 1.]]

y = [[ 0.], [ 1.]]

broadcast_equal(x, y) = [[ 0., 0., 0.], [ 1., 1., 1.]]

Defined in src/operator/tensor/elemwise_binary_broadcast_op_logic.cc:L46

Arguments

lhs::NDArray-or-SymbolicNode: First input to the function
rhs::NDArray-or-SymbolicNode: Second input to the function
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.broadcast_greater — Method.

broadcast_greater(lhs, rhs)

Returns the result of element-wise greater than (>) comparison operation with broadcasting.

Example::

x = [[ 1., 1., 1.], [ 1., 1., 1.]]

y = [[ 0.], [ 1.]]

broadcast_greater(x, y) = [[ 1., 1., 1.], [ 0., 0., 0.]]

Defined in src/operator/tensor/elemwise_binary_broadcast_op_logic.cc:L82

Arguments

lhs::NDArray-or-SymbolicNode: First input to the function
rhs::NDArray-or-SymbolicNode: Second input to the function
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.broadcast_greater_equal — Method.

broadcast_greater_equal(lhs, rhs)

Returns the result of element-wise greater than or equal to (>=) comparison operation with broadcasting.

Example::

x = [[ 1., 1., 1.], [ 1., 1., 1.]]

y = [[ 0.], [ 1.]]

broadcast_greater_equal(x, y) = [[ 1., 1., 1.], [ 1., 1., 1.]]

Defined in src/operator/tensor/elemwise_binary_broadcast_op_logic.cc:L100

Arguments

lhs::NDArray-or-SymbolicNode: First input to the function
rhs::NDArray-or-SymbolicNode: Second input to the function
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.broadcast_hypot — Method.

broadcast_hypot(lhs, rhs)

Returns the hypotenuse of a right angled triangle, given its "legs" with broadcasting.

It is equivalent to doing :math:sqrt(x_1^2 + x_2^2).

Example::

x = [[ 3., 3., 3.]]

y = [[ 4.], [ 4.]]

broadcast_hypot(x, y) = [[ 5., 5., 5.], [ 5., 5., 5.]]

z = [[ 0.], [ 4.]]

broadcast_hypot(x, z) = [[ 3., 3., 3.], [ 5., 5., 5.]]

Defined in src/operator/tensor/elemwise_binary_broadcast_op_extended.cc:L156

Arguments

lhs::NDArray-or-SymbolicNode: First input to the function
rhs::NDArray-or-SymbolicNode: Second input to the function
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.broadcast_lesser — Method.

broadcast_lesser(lhs, rhs)

Returns the result of element-wise lesser than (<) comparison operation with broadcasting.

Example::

x = [[ 1., 1., 1.], [ 1., 1., 1.]]

y = [[ 0.], [ 1.]]

broadcast_lesser(x, y) = [[ 0., 0., 0.], [ 0., 0., 0.]]

Defined in src/operator/tensor/elemwise_binary_broadcast_op_logic.cc:L118

Arguments

lhs::NDArray-or-SymbolicNode: First input to the function
rhs::NDArray-or-SymbolicNode: Second input to the function
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.broadcast_lesser_equal — Method.

broadcast_lesser_equal(lhs, rhs)

Returns the result of element-wise lesser than or equal to (<=) comparison operation with broadcasting.

Example::

x = [[ 1., 1., 1.], [ 1., 1., 1.]]

y = [[ 0.], [ 1.]]

broadcast_lesser_equal(x, y) = [[ 0., 0., 0.], [ 1., 1., 1.]]

Defined in src/operator/tensor/elemwise_binary_broadcast_op_logic.cc:L136

Arguments

lhs::NDArray-or-SymbolicNode: First input to the function
rhs::NDArray-or-SymbolicNode: Second input to the function
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.broadcast_maximum — Method.

broadcast_maximum(lhs, rhs)

Returns element-wise maximum of the input arrays with broadcasting.

This function compares two input arrays and returns a new array having the element-wise maxima.

Example::

x = [[ 1., 1., 1.], [ 1., 1., 1.]]

y = [[ 0.], [ 1.]]

broadcast_maximum(x, y) = [[ 1., 1., 1.], [ 1., 1., 1.]]

Defined in src/operator/tensor/elemwise_binary_broadcast_op_extended.cc:L80

Arguments

lhs::NDArray-or-SymbolicNode: First input to the function
rhs::NDArray-or-SymbolicNode: Second input to the function
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.broadcast_minimum — Method.

broadcast_minimum(lhs, rhs)

Returns element-wise minimum of the input arrays with broadcasting.

This function compares two input arrays and returns a new array having the element-wise minima.

Example::

x = [[ 1., 1., 1.], [ 1., 1., 1.]]

y = [[ 0.], [ 1.]]

broadcast_maximum(x, y) = [[ 0., 0., 0.], [ 1., 1., 1.]]

Defined in src/operator/tensor/elemwise_binary_broadcast_op_extended.cc:L115

Arguments

lhs::NDArray-or-SymbolicNode: First input to the function
rhs::NDArray-or-SymbolicNode: Second input to the function
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.broadcast_minus — Method.

broadcast_minus(lhs, rhs)

broadcast_minus is an alias of broadcast_sub.

Returns element-wise difference of the input arrays with broadcasting.

broadcast_minus is an alias to the function broadcast_sub.

Example::

x = [[ 1., 1., 1.], [ 1., 1., 1.]]

y = [[ 0.], [ 1.]]

broadcast_sub(x, y) = [[ 1., 1., 1.], [ 0., 0., 0.]]

broadcast_minus(x, y) = [[ 1., 1., 1.], [ 0., 0., 0.]]

Defined in src/operator/tensor/elemwise_binary_broadcast_op_basic.cc:L90

Arguments

lhs::NDArray-or-SymbolicNode: First input to the function
rhs::NDArray-or-SymbolicNode: Second input to the function
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.broadcast_mod — Method.

broadcast_mod(lhs, rhs)

Returns element-wise modulo of the input arrays with broadcasting.

Example::

x = [[ 8., 8., 8.], [ 8., 8., 8.]]

y = [[ 2.], [ 3.]]

broadcast_mod(x, y) = [[ 0., 0., 0.], [ 2., 2., 2.]]

Defined in src/operator/tensor/elemwise_binary_broadcast_op_basic.cc:L190

Arguments

lhs::NDArray-or-SymbolicNode: First input to the function
rhs::NDArray-or-SymbolicNode: Second input to the function
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.broadcast_mul — Method.

broadcast_mul(lhs, rhs)

Returns element-wise product of the input arrays with broadcasting.

Example::

x = [[ 1., 1., 1.], [ 1., 1., 1.]]

y = [[ 0.], [ 1.]]

broadcast_mul(x, y) = [[ 0., 0., 0.], [ 1., 1., 1.]]

Defined in src/operator/tensor/elemwise_binary_broadcast_op_basic.cc:L123

Arguments

lhs::NDArray-or-SymbolicNode: First input to the function
rhs::NDArray-or-SymbolicNode: Second input to the function
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.broadcast_not_equal — Method.

broadcast_not_equal(lhs, rhs)

Returns the result of element-wise not equal to (!=) comparison operation with broadcasting.

Example::

x = [[ 1., 1., 1.], [ 1., 1., 1.]]

y = [[ 0.], [ 1.]]

broadcast_not_equal(x, y) = [[ 1., 1., 1.], [ 0., 0., 0.]]

Defined in src/operator/tensor/elemwise_binary_broadcast_op_logic.cc:L64

Arguments

lhs::NDArray-or-SymbolicNode: First input to the function
rhs::NDArray-or-SymbolicNode: Second input to the function
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.broadcast_plus — Method.

broadcast_plus(lhs, rhs)

broadcast_plus is an alias of broadcast_add.

Returns element-wise sum of the input arrays with broadcasting.

broadcast_plus is an alias to the function broadcast_add.

Example::

x = [[ 1., 1., 1.], [ 1., 1., 1.]]

y = [[ 0.], [ 1.]]

broadcast_add(x, y) = [[ 1., 1., 1.], [ 2., 2., 2.]]

broadcast_plus(x, y) = [[ 1., 1., 1.], [ 2., 2., 2.]]

Defined in src/operator/tensor/elemwise_binary_broadcast_op_basic.cc:L51

Arguments

lhs::NDArray-or-SymbolicNode: First input to the function
rhs::NDArray-or-SymbolicNode: Second input to the function
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.broadcast_power — Method.

broadcast_power(lhs, rhs)

Returns result of first array elements raised to powers from second array, element-wise with broadcasting.

Example::

x = [[ 1., 1., 1.], [ 1., 1., 1.]]

y = [[ 0.], [ 1.]]

broadcast_power(x, y) = [[ 2., 2., 2.], [ 4., 4., 4.]]

Defined in src/operator/tensor/elemwise_binary_broadcast_op_extended.cc:L45

Arguments

lhs::NDArray-or-SymbolicNode: First input to the function
rhs::NDArray-or-SymbolicNode: Second input to the function
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.broadcast_sub — Method.

broadcast_sub(lhs, rhs)

Returns element-wise difference of the input arrays with broadcasting.

broadcast_minus is an alias to the function broadcast_sub.

Example::

x = [[ 1., 1., 1.], [ 1., 1., 1.]]

y = [[ 0.], [ 1.]]

broadcast_sub(x, y) = [[ 1., 1., 1.], [ 0., 0., 0.]]

broadcast_minus(x, y) = [[ 1., 1., 1.], [ 0., 0., 0.]]

Defined in src/operator/tensor/elemwise_binary_broadcast_op_basic.cc:L90

Arguments

lhs::NDArray-or-SymbolicNode: First input to the function
rhs::NDArray-or-SymbolicNode: Second input to the function
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.cast — Method.

cast(data, dtype)

cast is an alias of Cast.

Casts all elements of the input to a new type.

.. note:: $Cast$ is deprecated. Use $cast$ instead.

Example::

cast([0.9, 1.3], dtype='int32') = [0, 1] cast([1e20, 11.1], dtype='float16') = [inf, 11.09375] cast([300, 11.1, 10.9, -1, -3], dtype='uint8') = [44, 11, 10, 255, 253]

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L310

Arguments

data::NDArray-or-SymbolicNode: The input.
dtype::{'float16', 'float32', 'float64', 'int32', 'uint8'}, required: Output data type.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.cast_storage — Method.

cast_storage(data, stype)

Casts tensor storage type to the new type.

When an NDArray with default storage type is cast to csr or row_sparse storage, the result is compact, which means:

for csr, zero values will not be retained
for row_sparse, row slices of all zeros will not be retained

The storage type of $cast_storage$ output depends on stype parameter:

cast_storage(csr, 'default') = default
cast_storage(row_sparse, 'default') = default
cast_storage(default, 'csr') = csr
cast_storage(default, 'row_sparse') = row_sparse

Example::

dense = [[ 0.,  1.,  0.],
         [ 2.,  0.,  3.],
         [ 0.,  0.,  0.],
         [ 0.,  0.,  0.]]

# cast to row_sparse storage type
rsp = cast_storage(dense, 'row_sparse')
rsp.indices = [0, 1]
rsp.values = [[ 0.,  1.,  0.],
              [ 2.,  0.,  3.]]

# cast to csr storage type
csr = cast_storage(dense, 'csr')
csr.indices = [1, 0, 2]
csr.values = [ 1.,  2.,  3.]
csr.indptr = [0, 1, 3, 3, 3]

Defined in src/operator/tensor/cast_storage.cc:L69

Arguments

data::NDArray-or-SymbolicNode: The input.
stype::{'csr', 'default', 'row_sparse'}, required: Output storage type.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.choose_element_0index — Method.

choose_element_0index(lhs, rhs)

Choose one element from each line(row for python, column for R/Julia) in lhs according to index indicated by rhs. This function assume rhs uses 0-based index.

Arguments

lhs::NDArray: Left operand to the function.
rhs::NDArray: Right operand to the function.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.concat — Method.

concat(data, num_args, dim)

concat is an alias of Concat.

Note: concat takes variable number of positional inputs. So instead of calling as concat([x, y, z], num_args=3), one should call via concat(x, y, z), and num_args will be determined automatically.

Joins input arrays along a given axis.

.. note:: Concat is deprecated. Use concat instead.

The dimensions of the input arrays should be the same except the axis along which they will be concatenated. The dimension of the output array along the concatenated axis will be equal to the sum of the corresponding dimensions of the input arrays.

Example::

x = [[1,1],[2,2]] y = [[3,3],[4,4],[5,5]] z = [[6,6], [7,7],[8,8]]

concat(x,y,z,dim=0) = [[ 1., 1.], [ 2., 2.], [ 3., 3.], [ 4., 4.], [ 5., 5.], [ 6., 6.], [ 7., 7.], [ 8., 8.]]

Note that you cannot concat x,y,z along dimension 1 since dimension 0 is not the same for all the input arrays.

concat(y,z,dim=1) = [[ 3., 3., 6., 6.], [ 4., 4., 7., 7.], [ 5., 5., 8., 8.]]

Defined in src/operator/concat.cc:L104

Arguments

data::NDArray-or-SymbolicNode[]: List of arrays to concatenate
num_args::int, required: Number of inputs to be concated.
dim::int, optional, default='1': the dimension to be concated.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.crop — Method.

crop(data, begin, end, step)

crop is an alias of slice.

Slices a region of the array.

.. note:: $crop$ is deprecated. Use $slice$ instead.

This function returns a sliced array between the indices given by begin and end with the corresponding step.

For an input array of $shape=(d_0, d_1, ..., d_n-1)$, slice operation with $begin=(b_0, b_1...b_m-1)$, $end=(e_0, e_1, ..., e_m-1)$, and $step=(s_0, s_1, ..., s_m-1)$, where m <= n, results in an array with the shape $(|e_0-b_0|/|s_0|, ..., |e_m-1-b_m-1|/|s_m-1|, d_m, ..., d_n-1)$.

The resulting array's k-th dimension contains elements from the k-th dimension of the input array starting from index $b_k$ (inclusive) with step $s_k$ until reaching $e_k$ (exclusive).

If the k-th elements are None in the sequence of begin, end, and step, the following rule will be used to set default values. If s_k is None, set s_k=1. If s_k > 0, set b_k=0, e_k=d_k; else, set b_k=d_k-1, e_k=-1.

The storage type of $slice$ output depends on storage types of inputs

slice(csr) = csr
otherwise, $slice$ generates output with default storage

.. note:: When input data storage type is csr, it only supports step=(), or step=(None,), or step=(1,) to generate a csr output. For other step parameter values, it falls back to slicing a dense tensor.

Example::

x = [[ 1., 2., 3., 4.], [ 5., 6., 7., 8.], [ 9., 10., 11., 12.]]

slice(x, begin=(0,1), end=(2,4)) = [[ 2., 3., 4.], [ 6., 7., 8.]] slice(x, begin=(None, 0), end=(None, 3), step=(-1, 2)) = [[9., 11.], [5., 7.], [1., 3.]]

Defined in src/operator/tensor/matrix_op.cc:L359

Arguments

data::NDArray-or-SymbolicNode: Source input
begin::Shape(tuple), required: starting indices for the slice operation, supports negative indices.
end::Shape(tuple), required: ending indices for the slice operation, supports negative indices.
step::Shape(tuple), optional, default=[]: step for the slice operation, supports negative values.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.degrees — Method.

degrees(data)

Converts each element of the input array from radians to degrees.

.. math:: degrees([0, \pi/2, \pi, 3\pi/2, 2\pi]) = [0, 90, 180, 270, 360]

The storage type of $degrees$ output depends upon the input storage type:

degrees(default) = default
degrees(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_trig.cc:L163

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.elemwise_add — Method.

elemwise_add(lhs, rhs)

Adds arguments element-wise.

The storage type of $elemwise_add$ output depends on storage types of inputs

elemwise_add(row_sparse, row_sparse) = row_sparse
elemwise_add(csr, csr) = csr
otherwise, $elemwise_add$ generates output with default storage

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.elemwise_div — Method.

elemwise_div(lhs, rhs)

Divides arguments element-wise.

The storage type of $elemwise_div$ output is always dense

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.elemwise_mul — Method.

elemwise_mul(lhs, rhs)

Multiplies arguments element-wise.

The storage type of $elemwise_mul$ output depends on storage types of inputs

elemwise_mul(default, default) = default
elemwise_mul(row_sparse, row_sparse) = row_sparse
elemwise_mul(default, row_sparse) = default
elemwise_mul(row_sparse, default) = default
elemwise_mul(csr, csr) = csr
otherwise, $elemwise_mul$ generates output with default storage

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.elemwise_sub — Method.

elemwise_sub(lhs, rhs)

Subtracts arguments element-wise.

The storage type of $elemwise_sub$ output depends on storage types of inputs

elemwise_sub(row_sparse, row_sparse) = row_sparse
elemwise_sub(csr, csr) = csr
otherwise, $elemwise_sub$ generates output with default storage

Arguments

lhs::NDArray-or-SymbolicNode: first input
rhs::NDArray-or-SymbolicNode: second input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.fill_element_0index — Method.

fill_element_0index(lhs, mhs, rhs)

Fill one element of each line(row for python, column for R/Julia) in lhs according to index indicated by rhs and values indicated by mhs. This function assume rhs uses 0-based index.

Arguments

lhs::NDArray: Left operand to the function.
mhs::NDArray: Middle operand to the function.
rhs::NDArray: Right operand to the function.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.fix — Method.

fix(data)

Returns element-wise rounded value to the nearest integer towards zero of the input.

Example::

fix([-2.1, -1.9, 1.9, 2.1]) = [-2., -1., 1., 2.]

The storage type of $fix$ output depends upon the input storage type:

fix(default) = default
fix(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L520

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.flatten — Method.

flatten(data)

flatten is an alias of Flatten.

Flattens the input array into a 2-D array by collapsing the higher dimensions.

.. note:: Flatten is deprecated. Use flatten instead.

For an input array with shape $(d1, d2, ..., dk)$, flatten operation reshapes the input array into an output array of shape $(d1, d2...dk)$.

Note that the bahavior of this function is different from numpy.ndarray.flatten, which behaves similar to mxnet.ndarray.reshape((-1,)).

Example::

x = [[
    [1,2,3],
    [4,5,6],
    [7,8,9]
],
[    [1,2,3],
    [4,5,6],
    [7,8,9]
]],

flatten(x) = [[ 1.,  2.,  3.,  4.,  5.,  6.,  7.,  8.,  9.],
   [ 1.,  2.,  3.,  4.,  5.,  6.,  7.,  8.,  9.]]

Defined in src/operator/tensor/matrix_op.cc:L212

Arguments

data::NDArray-or-SymbolicNode: Input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.flip — Method.

flip(data, axis)

flip is an alias of reverse.

Reverses the order of elements along given axis while preserving array shape.

Note: reverse and flip are equivalent. We use reverse in the following examples.

Examples::

x = [[ 0., 1., 2., 3., 4.], [ 5., 6., 7., 8., 9.]]

reverse(x, axis=0) = [[ 5., 6., 7., 8., 9.], [ 0., 1., 2., 3., 4.]]

reverse(x, axis=1) = [[ 4., 3., 2., 1., 0.], [ 9., 8., 7., 6., 5.]]

Defined in src/operator/tensor/matrix_op.cc:L665

Arguments

data::NDArray-or-SymbolicNode: Input data array
axis::Shape(tuple), required: The axis which to reverse elements.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.from_json — Method.

from_json(repr :: AbstractString, ::Type{SymbolicNode})

Load a SymbolicNode from a JSON string representation.

source

# MXNet.mx.ftml_update — Method.

ftml_update(weight, grad, d, v, z, lr, beta1, beta2, epsilon, wd, rescale_grad, clip_gradient)

The FTML optimizer described in FTML - Follow the Moving Leader in Deep Learning, available at http://proceedings.mlr.press/v70/zheng17a/zheng17a.pdf.

.. math::

g_t = \nabla J(W_{t-1})\ v_t = \beta_2 v_{t-1} + (1 - \beta_2) g_t^2\ d_t = \frac{ (1 - \beta_1^t) }{ \eta_t } (\sqrt{ \frac{ v_t }{ 1 - \beta_2^t } } + \epsilon) \sigma_t = d_t - \beta_1 d_{t-1} z_t = \beta_1 z_{ t-1 } + (1 - \beta_1^t) g_t - \sigma_t W_{t-1} W_t = - \frac{ z_t }{ d_t }

Defined in src/operator/optimizer_op.cc:L336

Arguments

weight::NDArray-or-SymbolicNode: Weight
grad::NDArray-or-SymbolicNode: Gradient
d::NDArray-or-SymbolicNode: Internal state $d_t$
v::NDArray-or-SymbolicNode: Internal state $v_t$
z::NDArray-or-SymbolicNode: Internal state $z_t$
lr::float, required: Learning rate
beta1::float, optional, default=0.9: The decay rate for the 1st moment estimates.
beta2::float, optional, default=0.999: The decay rate for the 2nd moment estimates.
epsilon::float, optional, default=1e-08: A small constant for numerical stability.
wd::float, optional, default=0: Weight decay augments the objective function with a regularization term that penalizes large weights. The penalty scales with the square of the magnitude of each weight.
rescale_grad::float, optional, default=1: Rescale gradient to grad = rescale_grad*grad.
clip_gradient::float, optional, default=-1: Clip gradient to the range of [-clip_gradient, clip_gradient] If clip_gradient <= 0, gradient clipping is turned off. grad = max(min(grad, clip_gradient), -clip_gradient).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.ftrl_update — Method.

ftrl_update(weight, grad, z, n, lr, lamda1, beta, wd, rescale_grad, clip_gradient)

Update function for Ftrl optimizer. Referenced from Ad Click Prediction: a View from the Trenches, available at http://dl.acm.org/citation.cfm?id=2488200.

It updates the weights using::

rescaled_grad = clip(grad * rescale_grad, clip_gradient) z += rescaled_grad - (sqrt(n + rescaled_grad2) - sqrt(n)) * weight / learning_rate n += rescaled_grad2 w = (sign(z) * lamda1 - z) / ((beta + sqrt(n)) / learning_rate + wd) * (abs(z) > lamda1)

If w, z and n are all of $row_sparse$ storage type, only the row slices whose indices appear in grad.indices are updated (for w, z and n)::

for row in grad.indices: rescaled_grad[row] = clip(grad[row] * rescale_grad, clip_gradient) z[row] += rescaled_grad[row] - (sqrt(n[row] + rescaled_grad[row]2) - sqrt(n[row])) * weight[row] / learning_rate n[row] += rescaled_grad[row]2 w[row] = (sign(z[row]) * lamda1 - z[row]) / ((beta + sqrt(n[row])) / learning_rate + wd) * (abs(z[row]) > lamda1)

Defined in src/operator/optimizer_op.cc:L520

Arguments

weight::NDArray-or-SymbolicNode: Weight
grad::NDArray-or-SymbolicNode: Gradient
z::NDArray-or-SymbolicNode: z
n::NDArray-or-SymbolicNode: Square of grad
lr::float, required: Learning rate
lamda1::float, optional, default=0.01: The L1 regularization coefficient.
beta::float, optional, default=1: Per-Coordinate Learning Rate beta.
wd::float, optional, default=0: Weight decay augments the objective function with a regularization term that penalizes large weights. The penalty scales with the square of the magnitude of each weight.
rescale_grad::float, optional, default=1: Rescale gradient to grad = rescale_grad*grad.
clip_gradient::float, optional, default=-1: Clip gradient to the range of [-clip_gradient, clip_gradient] If clip_gradient <= 0, gradient clipping is turned off. grad = max(min(grad, clip_gradient), -clip_gradient).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.gammaln — Method.

gammaln(data)

Returns element-wise log of the absolute value of the gamma function of the input.

The storage type of $gammaln$ output is always dense

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.gather_nd — Method.

gather_nd(data, indices)

Gather elements or slices from data and store to a tensor whose shape is defined by indices.

Given data with shape (X_0, X_1, ..., X_{N-1}) and indices with shape (M, Y_0, ..., Y_{K-1}), the output will have shape (Y_0, ..., Y_{K-1}, X_M, ..., X_{N-1}), where M <= N. If M == N, output shape will simply be (Y_0, ..., Y_{K-1}).

The elements in output is defined as follows::

output[y_0, ..., y_{K-1}, x_M, ..., x_{N-1}] = data[indices[0, y_0, ..., y_{K-1}], ..., indices[M-1, y_0, ..., y_{K-1}], x_M, ..., x_{N-1}]

Examples::

data = [[0, 1], [2, 3]] indices = [[1, 1, 0], [0, 1, 0]] gather_nd(data, indices) = [2, 3, 0]

Arguments

data::NDArray-or-SymbolicNode: data
indices::NDArray-or-SymbolicNode: indices
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.get_attr — Method.

get_attr(self :: SymbolicNode, key :: Symbol)

Get attribute attached to this SymbolicNode belonging to key.

Returns the value belonging to key as a Nullable.

source

# MXNet.mx.get_children — Method.

get_children(x::SymbolicNode)

Gets a new grouped SymbolicNode whose output contains inputs to output nodes of the original symbol.

julia> x = mx.Variable(:x)
MXNet.mx.SymbolicNode x

julia> y = mx.Variable(:y)
MXNet.mx.SymbolicNode y

julia> z = x + y
MXNet.mx.SymbolicNode _plus1

julia> a |> mx.get_children |> mx.list_outputs
2-element Array{Symbol,1}:
 :x
 :y

source

# MXNet.mx.get_internals — Method.

get_internals(self :: SymbolicNode)

Get a new grouped SymbolicNode whose output contains all the internal outputs of this SymbolicNode.

source

# MXNet.mx.get_name — Method.

get_name(self :: SymbolicNode)

Get the name of the symbol.

julia> x = mx.Variable(:data)
julia> mx.get_name(x)
:data

julia> y = mx.FullyConnected(x, num_hidden = 128)
julia> mx.get_name(y)
:fullyconnected0

source

# MXNet.mx.grad — Method.

grad(self :: SymbolicNode, wrt :: Vector{SymbolicNode})

Get the autodiff gradient of the current SymbolicNode. This function can only be used if the current symbol is a loss function.

Arguments:

self::SymbolicNode: current node.
wrt::Vector{Symbol}: the names of the arguments to the gradient.

Returns a gradient symbol of the corresponding gradient.

source

# MXNet.mx.infer_shape — Method.

infer_shape(self :: SymbolicNode, args...)
infer_shape(self :: SymbolicNode; kwargs...)

Do shape inference according to the input shapes. The input shapes could be provided as a list of shapes, which should specify the shapes of inputs in the same order as the arguments returned by list_arguments. Alternatively, the shape information could be specified via keyword arguments.

Returns a 3-tuple containing shapes of all the arguments, shapes of all the outputs and shapes of all the auxiliary variables. If shape inference failed due to incomplete or incompatible inputs, the return value will be (nothing, nothing, nothing).

source

# MXNet.mx.infer_type — Method.

infer_type(self :: SymbolicNode; kwargs...)
infer_type(self :: SymbolicNode, args...)

Do type inference according to the input types. The input types could be provided as a list of types, which should specify the types of inputs in the same order as the arguments returned by list_arguments. Alternatively, the type information could be specified via keyword arguments.

Returns a 3-tuple containing types of all the arguments, types of all the outputs and types of all the auxiliary variables. If type inference failed due to incomplete or incompatible inputs, the return value will be (nothing, nothing, nothing).

source

# MXNet.mx.khatri_rao — Method.

khatri_rao(args)

Note: khatri_rao takes variable number of positional inputs. So instead of calling as khatri_rao([x, y, z], num_args=3), one should call via khatri_rao(x, y, z), and num_args will be determined automatically.

Computes the Khatri-Rao product of the input matrices.

Given a collection of :math:n input matrices,

.. math:: A_1 \in \mathbb{R}^{M_1 \times M}, \ldots, A_n \in \mathbb{R}^{M_n \times N},

the (column-wise) Khatri-Rao product is defined as the matrix,

.. math:: X = A_1 \otimes \cdots \otimes A_n \in \mathbb{R}^{(M_1 \cdots M_n) \times N},

where the :math:kth column is equal to the column-wise outer product :math:{A_1}_k \otimes \cdots \otimes {A_n}_k where :math:{A_i}_k is the kth column of the ith matrix.

Example::

A = mx.nd.array([[1, -1], [2, -3]]) B = mx.nd.array([[1, 4], [2, 5], [3, 6]]) C = mx.nd.khatri_rao(A, B) print(C.asnumpy())

[[ 1. -4.] [ 2. -5.] [ 3. -6.] [ 2. -12.] [ 4. -15.] [ 6. -18.]]

Defined in src/operator/contrib/krprod.cc:L108

Arguments

args::NDArray-or-SymbolicNode[]: Positional input matrices
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.linalg_gelqf — Method.

linalg_gelqf(A)

linalg_gelqf is an alias of _linalg_gelqf.

LQ factorization for general matrix. Input is a tensor A of dimension n >= 2.

If n=2, we compute the LQ factorization (LAPACK gelqf, followed by orglq). A must have shape (x, y) with x <= y, and must have full rank =x. The LQ factorization consists of L with shape (x, x) and Q with shape (x, y), so that:

A = L * Q

Here, L is lower triangular (upper triangle equal to zero) with nonzero diagonal, and Q is row-orthonormal, meaning that

Q * Q\ :sup:T

is equal to the identity matrix of shape (x, x).

If n>2, gelqf is performed separately on the trailing two dimensions for all inputs (batch mode).

.. note:: The operator supports float32 and float64 data types only.

Examples::

// Single LQ factorization A = [[1., 2., 3.], [4., 5., 6.]] Q, L = gelqf(A) Q = [[-0.26726124, -0.53452248, -0.80178373], [0.87287156, 0.21821789, -0.43643578]] L = [[-3.74165739, 0.], [-8.55235974, 1.96396101]]

// Batch LQ factorization A = [[[1., 2., 3.], [4., 5., 6.]], [[7., 8., 9.], [10., 11., 12.]]] Q, L = gelqf(A) Q = [[[-0.26726124, -0.53452248, -0.80178373], [0.87287156, 0.21821789, -0.43643578]], [[-0.50257071, -0.57436653, -0.64616234], [0.7620735, 0.05862104, -0.64483142]]] L = [[[-3.74165739, 0.], [-8.55235974, 1.96396101]], [[-13.92838828, 0.], [-19.09768702, 0.52758934]]]

Defined in src/operator/tensor/la_op.cc:L529

Arguments

A::NDArray-or-SymbolicNode: Tensor of input matrices to be factorized
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.linalg_gemm — Method.

linalg_gemm(A, B, C, transpose_a, transpose_b, alpha, beta)

linalg_gemm is an alias of _linalg_gemm.

Performs general matrix multiplication and accumulation. Input are tensors A, B, C, each of dimension n >= 2 and having the same shape on the leading n-2 dimensions.

If n=2, the BLAS3 function gemm is performed:

out = alpha * op\ (A) * op\ (B) + beta * C

Here, alpha and beta are scalar parameters, and op() is either the identity or matrix transposition (depending on transpose_a, transpose_b).

If n>2, gemm is performed separately on the trailing two dimensions for all inputs (batch mode).

.. note:: The operator supports float32 and float64 data types only.

Examples::

// Single matrix multiply-add A = [[1.0, 1.0], [1.0, 1.0]] B = [[1.0, 1.0], [1.0, 1.0], [1.0, 1.0]] C = [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0]] gemm(A, B, C, transpose_b=True, alpha=2.0, beta=10.0) = [[14.0, 14.0, 14.0], [14.0, 14.0, 14.0]]

// Batch matrix multiply-add A = [[[1.0, 1.0]], [[0.1, 0.1]]] B = [[[1.0, 1.0]], [[0.1, 0.1]]] C = [[[10.0]], [[0.01]]] gemm(A, B, C, transpose_b=True, alpha=2.0 , beta=10.0) = [[[104.0]], [[0.14]]]

Defined in src/operator/tensor/la_op.cc:L69

Arguments

A::NDArray-or-SymbolicNode: Tensor of input matrices
B::NDArray-or-SymbolicNode: Tensor of input matrices
C::NDArray-or-SymbolicNode: Tensor of input matrices
transpose_a::boolean, optional, default=0: Multiply with transposed of first input (A).
transpose_b::boolean, optional, default=0: Multiply with transposed of second input (B).
alpha::double, optional, default=1: Scalar factor multiplied with A*B.
beta::double, optional, default=1: Scalar factor multiplied with C.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.linalg_gemm2 — Method.

linalg_gemm2(A, B, transpose_a, transpose_b, alpha)

linalg_gemm2 is an alias of _linalg_gemm2.

Performs general matrix multiplication. Input are tensors A, B, each of dimension n >= 2 and having the same shape on the leading n-2 dimensions.

If n=2, the BLAS3 function gemm is performed:

out = alpha * op\ (A) * op\ (B)

Here alpha is a scalar parameter and op() is either the identity or the matrix transposition (depending on transpose_a, transpose_b).

If n>2, gemm is performed separately on the trailing two dimensions for all inputs (batch mode).

.. note:: The operator supports float32 and float64 data types only.

Examples::

// Single matrix multiply A = [[1.0, 1.0], [1.0, 1.0]] B = [[1.0, 1.0], [1.0, 1.0], [1.0, 1.0]] gemm2(A, B, transpose_b=True, alpha=2.0) = [[4.0, 4.0, 4.0], [4.0, 4.0, 4.0]]

// Batch matrix multiply A = [[[1.0, 1.0]], [[0.1, 0.1]]] B = [[[1.0, 1.0]], [[0.1, 0.1]]] gemm2(A, B, transpose_b=True, alpha=2.0) = [[[4.0]], [[0.04 ]]]

Defined in src/operator/tensor/la_op.cc:L128

Arguments

A::NDArray-or-SymbolicNode: Tensor of input matrices
B::NDArray-or-SymbolicNode: Tensor of input matrices
transpose_a::boolean, optional, default=0: Multiply with transposed of first input (A).
transpose_b::boolean, optional, default=0: Multiply with transposed of second input (B).
alpha::double, optional, default=1: Scalar factor multiplied with A*B.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.linalg_potrf — Method.

linalg_potrf(A)

linalg_potrf is an alias of _linalg_potrf.

Performs Cholesky factorization of a symmetric positive-definite matrix. Input is a tensor A of dimension n >= 2.

If n=2, the Cholesky factor L of the symmetric, positive definite matrix A is computed. L is lower triangular (entries of upper triangle are all zero), has positive diagonal entries, and:

A = L * L\ :sup:T

If n>2, potrf is performed separately on the trailing two dimensions for all inputs (batch mode).

.. note:: The operator supports float32 and float64 data types only.

Examples::

// Single matrix factorization A = [[4.0, 1.0], [1.0, 4.25]] potrf(A) = [[2.0, 0], [0.5, 2.0]]

// Batch matrix factorization A = [[[4.0, 1.0], [1.0, 4.25]], [[16.0, 4.0], [4.0, 17.0]]] potrf(A) = [[[2.0, 0], [0.5, 2.0]], [[4.0, 0], [1.0, 4.0]]]

Defined in src/operator/tensor/la_op.cc:L178

Arguments

A::NDArray-or-SymbolicNode: Tensor of input matrices to be decomposed
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.linalg_potri — Method.

linalg_potri(A)

linalg_potri is an alias of _linalg_potri.

Performs matrix inversion from a Cholesky factorization. Input is a tensor A of dimension n >= 2.

If n=2, A is a lower triangular matrix (entries of upper triangle are all zero) with positive diagonal. We compute:

out = A\ :sup:-T * A\ :sup:-1

In other words, if A is the Cholesky factor of a symmetric positive definite matrix B (obtained by potrf), then

out = B\ :sup:-1

If n>2, potri is performed separately on the trailing two dimensions for all inputs (batch mode).

.. note:: The operator supports float32 and float64 data types only.

.. note:: Use this operator only if you are certain you need the inverse of B, and cannot use the Cholesky factor A (potrf), together with backsubstitution (trsm). The latter is numerically much safer, and also cheaper.

Examples::

// Single matrix inverse A = [[2.0, 0], [0.5, 2.0]] potri(A) = [[0.26563, -0.0625], [-0.0625, 0.25]]

// Batch matrix inverse A = [[[2.0, 0], [0.5, 2.0]], [[4.0, 0], [1.0, 4.0]]] potri(A) = [[[0.26563, -0.0625], [-0.0625, 0.25]], [[0.06641, -0.01562], [-0.01562, 0,0625]]]

Defined in src/operator/tensor/la_op.cc:L236

Arguments

A::NDArray-or-SymbolicNode: Tensor of lower triangular matrices
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.linalg_sumlogdiag — Method.

linalg_sumlogdiag(A)

linalg_sumlogdiag is an alias of _linalg_sumlogdiag.

Computes the sum of the logarithms of the diagonal elements of a square matrix. Input is a tensor A of dimension n >= 2.

If n=2, A must be square with positive diagonal entries. We sum the natural logarithms of the diagonal elements, the result has shape (1,).

If n>2, sumlogdiag is performed separately on the trailing two dimensions for all inputs (batch mode).

.. note:: The operator supports float32 and float64 data types only.

Examples::

// Single matrix reduction A = [[1.0, 1.0], [1.0, 7.0]] sumlogdiag(A) = [1.9459]

// Batch matrix reduction A = [[[1.0, 1.0], [1.0, 7.0]], [[3.0, 0], [0, 17.0]]] sumlogdiag(A) = [1.9459, 3.9318]

Defined in src/operator/tensor/la_op.cc:L405

Arguments

A::NDArray-or-SymbolicNode: Tensor of square matrices
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.linalg_syrk — Method.

linalg_syrk(A, transpose, alpha)

linalg_syrk is an alias of _linalg_syrk.

Multiplication of matrix with its transpose. Input is a tensor A of dimension n >= 2.

If n=2, the operator performs the BLAS3 function syrk:

out = alpha * A * A\ :sup:T

if transpose=False, or

out = alpha * A\ :sup:T \ * A

if transpose=True.

If n>2, syrk is performed separately on the trailing two dimensions for all inputs (batch mode).

.. note:: The operator supports float32 and float64 data types only.

Examples::

// Single matrix multiply A = [[1., 2., 3.], [4., 5., 6.]] syrk(A, alpha=1., transpose=False) = [[14., 32.], [32., 77.]] syrk(A, alpha=1., transpose=True) = [[17., 22., 27.], [22., 29., 36.], [27., 36., 45.]]

// Batch matrix multiply A = [[[1., 1.]], [[0.1, 0.1]]] syrk(A, alpha=2., transpose=False) = [[[4.]], [[0.04]]]

Defined in src/operator/tensor/la_op.cc:L461

Arguments

A::NDArray-or-SymbolicNode: Tensor of input matrices
transpose::boolean, optional, default=0: Use transpose of input matrix.
alpha::double, optional, default=1: Scalar factor to be applied to the result.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.linalg_trmm — Method.

linalg_trmm(A, B, transpose, rightside, alpha)

linalg_trmm is an alias of _linalg_trmm.

Performs multiplication with a lower triangular matrix. Input are tensors A, B, each of dimension n >= 2 and having the same shape on the leading n-2 dimensions.

If n=2, A must be lower triangular. The operator performs the BLAS3 function trmm:

out = alpha * op\ (A) * B

if rightside=False, or

out = alpha * B * op\ (A)

if rightside=True. Here, alpha is a scalar parameter, and op() is either the identity or the matrix transposition (depending on transpose).

If n>2, trmm is performed separately on the trailing two dimensions for all inputs (batch mode).

.. note:: The operator supports float32 and float64 data types only.

Examples::

// Single triangular matrix multiply A = [[1.0, 0], [1.0, 1.0]] B = [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0]] trmm(A, B, alpha=2.0) = [[2.0, 2.0, 2.0], [4.0, 4.0, 4.0]]

// Batch triangular matrix multiply A = [[[1.0, 0], [1.0, 1.0]], [[1.0, 0], [1.0, 1.0]]] B = [[[1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], [[0.5, 0.5, 0.5], [0.5, 0.5, 0.5]]] trmm(A, B, alpha=2.0) = [[[2.0, 2.0, 2.0], [4.0, 4.0, 4.0]], [[1.0, 1.0, 1.0], [2.0, 2.0, 2.0]]]

Defined in src/operator/tensor/la_op.cc:L293

Arguments

A::NDArray-or-SymbolicNode: Tensor of lower triangular matrices
B::NDArray-or-SymbolicNode: Tensor of matrices
transpose::boolean, optional, default=0: Use transposed of the triangular matrix
rightside::boolean, optional, default=0: Multiply triangular matrix from the right to non-triangular one.
alpha::double, optional, default=1: Scalar factor to be applied to the result.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.linalg_trsm — Method.

linalg_trsm(A, B, transpose, rightside, alpha)

linalg_trsm is an alias of _linalg_trsm.

Solves matrix equation involving a lower triangular matrix. Input are tensors A, B, each of dimension n >= 2 and having the same shape on the leading n-2 dimensions.

If n=2, A must be lower triangular. The operator performs the BLAS3 function trsm, solving for out in:

op\ (A) * out = alpha * B

if rightside=False, or

out * op\ (A) = alpha * B

if rightside=True. Here, alpha is a scalar parameter, and op() is either the identity or the matrix transposition (depending on transpose).

If n>2, trsm is performed separately on the trailing two dimensions for all inputs (batch mode).

.. note:: The operator supports float32 and float64 data types only.

Examples::

// Single matrix solve A = [[1.0, 0], [1.0, 1.0]] B = [[2.0, 2.0, 2.0], [4.0, 4.0, 4.0]] trsm(A, B, alpha=0.5) = [[1.0, 1.0, 1.0], [1.0, 1.0, 1.0]]

// Batch matrix solve A = [[[1.0, 0], [1.0, 1.0]], [[1.0, 0], [1.0, 1.0]]] B = [[[2.0, 2.0, 2.0], [4.0, 4.0, 4.0]], [[4.0, 4.0, 4.0], [8.0, 8.0, 8.0]]] trsm(A, B, alpha=0.5) = [[[1.0, 1.0, 1.0], [1.0, 1.0, 1.0]], [[2.0, 2.0, 2.0], [2.0, 2.0, 2.0]]]

Defined in src/operator/tensor/la_op.cc:L356

Arguments

A::NDArray-or-SymbolicNode: Tensor of lower triangular matrices
B::NDArray-or-SymbolicNode: Tensor of matrices
transpose::boolean, optional, default=0: Use transposed of the triangular matrix
rightside::boolean, optional, default=0: Multiply triangular matrix from the right to non-triangular one.
alpha::double, optional, default=1: Scalar factor to be applied to the result.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.list_all_attr — Method.

list_all_attr(self :: SymbolicNode)

Get all attributes from the symbol graph.

Returns a dictionary of attributes.

source

# MXNet.mx.list_arguments — Method.

list_arguments(self :: SymbolicNode)

List all the arguments of this node. The argument for a node contains both the inputs and parameters. For example, a FullyConnected node will have both data and weights in its arguments. A composed node (e.g. a MLP) will list all the arguments for intermediate nodes.

Returns a list of symbols indicating the names of the arguments.

source

# MXNet.mx.list_attr — Method.

list_attr(self :: SymbolicNode)

Get all attributes from a symbol.

Returns a dictionary of attributes.

source

# MXNet.mx.list_auxiliary_states — Method.

list_auxiliary_states(self :: SymbolicNode)

List all auxiliary states in the symbool.

Auxiliary states are special states of symbols that do not corresponds to an argument, and do not have gradient. But still be useful for the specific operations. A common example of auxiliary state is the moving_mean and moving_variance in BatchNorm. Most operators do not have Auxiliary states.

Returns a list of symbols indicating the names of the auxiliary states.

source

# MXNet.mx.list_outputs — Method.

list_outputs(self :: SymbolicNode)

List all the outputs of this node.

Returns a list of symbols indicating the names of the outputs.

source

# MXNet.mx.load — Method.

load(filename :: AbstractString, ::Type{SymbolicNode})

Load a SymbolicNode from a JSON file.

source

# MXNet.mx.make_loss — Method.

make_loss(data)

Make your own loss function in network construction.

This operator accepts a customized loss function symbol as a terminal loss and the symbol should be an operator with no backward dependency. The output of this function is the gradient of loss with respect to the input data.

For example, if you are a making a cross entropy loss function. Assume $out$ is the predicted output and $label$ is the true label, then the cross entropy can be defined as::

cross_entropy = label * log(out) + (1 - label) * log(1 - out) loss = make_loss(cross_entropy)

We will need to use $make_loss$ when we are creating our own loss function or we want to combine multiple loss functions. Also we may want to stop some variables' gradients from backpropagation. See more detail in $BlockGrad$ or $stop_gradient$.

The storage type of $make_loss$ output depends upon the input storage type:

make_loss(default) = default
make_loss(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L199

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.max_axis — Method.

max_axis(data, axis, keepdims, exclude)

max_axis is an alias of max.

Computes the max of array elements over given axes.

Defined in src/operator/tensor/broadcast_reduce_op_value.cc:L160

Arguments

data::NDArray-or-SymbolicNode: The input
axis::Shape(tuple), optional, default=[]: The axis or axes along which to perform the reduction.

`` The default,axis=(), will compute over all elements into a scalar array with shape(1,)`.

If axis is int, a reduction is performed on a particular axis.

If axis is a tuple of ints, a reduction is performed on all the axes specified in the tuple.

If exclude is true, reduction will be performed on the axes that are NOT in axis instead.

Negative values means indexing from right to left. `` *keepdims::boolean, optional, default=0: If this is set toTrue, the reduced axes are left in the result as dimension with size one. *exclude::boolean, optional, default=0: Whether to perform reduction on axis that are NOT in axis instead. *name::Symbol: The name of theSymbolicNode. (e.g.:my_symbol), optional. *attrs::Dict{Symbol, AbstractString}: The attributes associated with thisSymbolicNode`.

source

# MXNet.mx.min_axis — Method.

min_axis(data, axis, keepdims, exclude)

min_axis is an alias of min.

Computes the min of array elements over given axes.

Defined in src/operator/tensor/broadcast_reduce_op_value.cc:L174

Arguments

data::NDArray-or-SymbolicNode: The input
axis::Shape(tuple), optional, default=[]: The axis or axes along which to perform the reduction.

`` The default,axis=(), will compute over all elements into a scalar array with shape(1,)`.

If axis is int, a reduction is performed on a particular axis.

If axis is a tuple of ints, a reduction is performed on all the axes specified in the tuple.

If exclude is true, reduction will be performed on the axes that are NOT in axis instead.

Negative values means indexing from right to left. `` *keepdims::boolean, optional, default=0: If this is set toTrue, the reduced axes are left in the result as dimension with size one. *exclude::boolean, optional, default=0: Whether to perform reduction on axis that are NOT in axis instead. *name::Symbol: The name of theSymbolicNode. (e.g.:my_symbol), optional. *attrs::Dict{Symbol, AbstractString}: The attributes associated with thisSymbolicNode`.

source

# MXNet.mx.mp_sgd_mom_update — Method.

mp_sgd_mom_update(weight, grad, mom, weight32, lr, momentum, wd, rescale_grad, clip_gradient)

Updater function for multi-precision sgd optimizer

Arguments

weight::NDArray-or-SymbolicNode: Weight
grad::NDArray-or-SymbolicNode: Gradient
mom::NDArray-or-SymbolicNode: Momentum
weight32::NDArray-or-SymbolicNode: Weight32
lr::float, required: Learning rate
momentum::float, optional, default=0: The decay rate of momentum estimates at each epoch.
wd::float, optional, default=0: Weight decay augments the objective function with a regularization term that penalizes large weights. The penalty scales with the square of the magnitude of each weight.
rescale_grad::float, optional, default=1: Rescale gradient to grad = rescale_grad*grad.
clip_gradient::float, optional, default=-1: Clip gradient to the range of [-clip_gradient, clip_gradient] If clip_gradient <= 0, gradient clipping is turned off. grad = max(min(grad, clip_gradient), -clip_gradient).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.mp_sgd_update — Method.

mp_sgd_update(weight, grad, weight32, lr, wd, rescale_grad, clip_gradient)

Updater function for multi-precision sgd optimizer

Arguments

weight::NDArray-or-SymbolicNode: Weight
grad::NDArray-or-SymbolicNode: gradient
weight32::NDArray-or-SymbolicNode: Weight32
lr::float, required: Learning rate
wd::float, optional, default=0: Weight decay augments the objective function with a regularization term that penalizes large weights. The penalty scales with the square of the magnitude of each weight.
rescale_grad::float, optional, default=1: Rescale gradient to grad = rescale_grad*grad.
clip_gradient::float, optional, default=-1: Clip gradient to the range of [-clip_gradient, clip_gradient] If clip_gradient <= 0, gradient clipping is turned off. grad = max(min(grad, clip_gradient), -clip_gradient).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.nanprod — Method.

nanprod(data, axis, keepdims, exclude)

Computes the product of array elements over given axes treating Not a Numbers ($NaN$) as one.

Defined in src/operator/tensor/broadcast_reduce_op_value.cc:L146

Arguments

data::NDArray-or-SymbolicNode: The input
axis::Shape(tuple), optional, default=[]: The axis or axes along which to perform the reduction.

`` The default,axis=(), will compute over all elements into a scalar array with shape(1,)`.

If axis is int, a reduction is performed on a particular axis.

If axis is a tuple of ints, a reduction is performed on all the axes specified in the tuple.

If exclude is true, reduction will be performed on the axes that are NOT in axis instead.

Negative values means indexing from right to left. `` *keepdims::boolean, optional, default=0: If this is set toTrue, the reduced axes are left in the result as dimension with size one. *exclude::boolean, optional, default=0: Whether to perform reduction on axis that are NOT in axis instead. *name::Symbol: The name of theSymbolicNode. (e.g.:my_symbol), optional. *attrs::Dict{Symbol, AbstractString}: The attributes associated with thisSymbolicNode`.

source

# MXNet.mx.nansum — Method.

nansum(data, axis, keepdims, exclude)

Computes the sum of array elements over given axes treating Not a Numbers ($NaN$) as zero.

Defined in src/operator/tensor/broadcast_reduce_op_value.cc:L131

Arguments

data::NDArray-or-SymbolicNode: The input
axis::Shape(tuple), optional, default=[]: The axis or axes along which to perform the reduction.

`` The default,axis=(), will compute over all elements into a scalar array with shape(1,)`.

If axis is int, a reduction is performed on a particular axis.

If axis is a tuple of ints, a reduction is performed on all the axes specified in the tuple.

If exclude is true, reduction will be performed on the axes that are NOT in axis instead.

Negative values means indexing from right to left. `` *keepdims::boolean, optional, default=0: If this is set toTrue, the reduced axes are left in the result as dimension with size one. *exclude::boolean, optional, default=0: Whether to perform reduction on axis that are NOT in axis instead. *name::Symbol: The name of theSymbolicNode. (e.g.:my_symbol), optional. *attrs::Dict{Symbol, AbstractString}: The attributes associated with thisSymbolicNode`.

source

# MXNet.mx.negative — Method.

negative(data)

Numerical negative of the argument, element-wise.

The storage type of $negative$ output depends upon the input storage type:

negative(default) = default
negative(row_sparse) = row_sparse
negative(csr) = csr

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.normal — Method.

normal(loc, scale, shape, ctx, dtype)

normal is an alias of _random_normal.

Draw random samples from a normal (Gaussian) distribution.

.. note:: The existing alias $normal$ is deprecated.

Samples are distributed according to a normal distribution parametrized by loc (mean) and scale (standard deviation).

Example::

normal(loc=0, scale=1, shape=(2,2)) = [[ 1.89171135, -1.16881478], [-1.23474145, 1.55807114]]

Defined in src/operator/random/sample_op.cc:L85

Arguments

loc::float, optional, default=0: Mean of the distribution.
scale::float, optional, default=1: Standard deviation of the distribution.
shape::Shape(tuple), optional, default=[]: Shape of the output.
ctx::string, optional, default='': Context of output, in format cpu|gpu|cpu_pinned. Only used for imperative calls.
dtype::{'None', 'float16', 'float32', 'float64'},optional, default='None': DType of the output in case this can't be inferred. Defaults to float32 if not defined (dtype=None).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.one_hot — Method.

one_hot(indices, depth, on_value, off_value, dtype)

Returns a one-hot array.

The locations represented by indices take value on_value, while all other locations take value off_value.

one_hot operation with indices of shape $(i0, i1)$ and depth of $d$ would result in an output array of shape $(i0, i1, d)$ with::

output[i,j,:] = off_value output[i,j,indices[i,j]] = on_value

Examples::

one_hot([1,0,2,0], 3) = [[ 0. 1. 0.] [ 1. 0. 0.] [ 0. 0. 1.] [ 1. 0. 0.]]

one_hot([1,0,2,0], 3, on_value=8, off_value=1, dtype='int32') = [[1 8 1] [8 1 1] [1 1 8] [8 1 1]]

one_hot([[1,0],[1,0],[2,0]], 3) = [[[ 0. 1. 0.] [ 1. 0. 0.]]

                                 [[ 0.  1.  0.]
                                  [ 1.  0.  0.]]

                                 [[ 0.  0.  1.]
                                  [ 1.  0.  0.]]]

Defined in src/operator/tensor/indexing_op.cc:L472

Arguments

indices::NDArray-or-SymbolicNode: array of locations where to set on_value
depth::int, required: Depth of the one hot dimension.
on_value::double, optional, default=1: The value assigned to the locations represented by indices.
off_value::double, optional, default=0: The value assigned to the locations not represented by indices.
dtype::{'float16', 'float32', 'float64', 'int32', 'uint8'},optional, default='float32': DType of the output
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.ones_like — Method.

ones_like(data)

Return an array of ones with the same shape and type as the input array.

Examples::

x = [[ 0., 0., 0.], [ 0., 0., 0.]]

ones_like(x) = [[ 1., 1., 1.], [ 1., 1., 1.]]

Arguments

data::NDArray-or-SymbolicNode: The input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.pad — Method.

pad(data, mode, pad_width, constant_value)

pad is an alias of Pad.

Pads an input array with a constant or edge values of the array.

.. note:: Pad is deprecated. Use pad instead.

.. note:: Current implementation only supports 4D and 5D input arrays with padding applied only on axes 1, 2 and 3. Expects axes 4 and 5 in pad_width to be zero.

This operation pads an input array with either a constant_value or edge values along each axis of the input array. The amount of padding is specified by pad_width.

pad_width is a tuple of integer padding widths for each axis of the format $(before_1, after_1, ... , before_N, after_N)$. The pad_width should be of length $2*N$ where $N$ is the number of dimensions of the array.

For dimension $N$ of the input array, $before_N$ and $after_N$ indicates how many values to add before and after the elements of the array along dimension $N$. The widths of the higher two dimensions $before_1$, $after_1$, $before_2$, $after_2$ must be 0.

Example::

x = [[[[ 1. 2. 3.] [ 4. 5. 6.]]

     [[  7.   8.   9.]
      [ 10.  11.  12.]]]


    [[[ 11.  12.  13.]
      [ 14.  15.  16.]]

     [[ 17.  18.  19.]
      [ 20.  21.  22.]]]]

pad(x,mode="edge", pad_width=(0,0,0,0,1,1,1,1)) =

     [[[[  1.   1.   2.   3.   3.]
        [  1.   1.   2.   3.   3.]
        [  4.   4.   5.   6.   6.]
        [  4.   4.   5.   6.   6.]]

       [[  7.   7.   8.   9.   9.]
        [  7.   7.   8.   9.   9.]
        [ 10.  10.  11.  12.  12.]
        [ 10.  10.  11.  12.  12.]]]


      [[[ 11.  11.  12.  13.  13.]
        [ 11.  11.  12.  13.  13.]
        [ 14.  14.  15.  16.  16.]
        [ 14.  14.  15.  16.  16.]]

       [[ 17.  17.  18.  19.  19.]
        [ 17.  17.  18.  19.  19.]
        [ 20.  20.  21.  22.  22.]
        [ 20.  20.  21.  22.  22.]]]]

pad(x, mode="constant", constant_value=0, pad_width=(0,0,0,0,1,1,1,1)) =

     [[[[  0.   0.   0.   0.   0.]
        [  0.   1.   2.   3.   0.]
        [  0.   4.   5.   6.   0.]
        [  0.   0.   0.   0.   0.]]

       [[  0.   0.   0.   0.   0.]
        [  0.   7.   8.   9.   0.]
        [  0.  10.  11.  12.   0.]
        [  0.   0.   0.   0.   0.]]]


      [[[  0.   0.   0.   0.   0.]
        [  0.  11.  12.  13.   0.]
        [  0.  14.  15.  16.   0.]
        [  0.   0.   0.   0.   0.]]

       [[  0.   0.   0.   0.   0.]
        [  0.  17.  18.  19.   0.]
        [  0.  20.  21.  22.   0.]
        [  0.   0.   0.   0.   0.]]]]

Defined in src/operator/pad.cc:L766

Arguments

data::NDArray-or-SymbolicNode: An n-dimensional input array.
mode::{'constant', 'edge', 'reflect'}, required: Padding type to use. "constant" pads with constant_value "edge" pads using the edge values of the input array "reflect" pads by reflecting values with respect to the edges.
pad_width::Shape(tuple), required: Widths of the padding regions applied to the edges of each axis. It is a tuple of integer padding widths for each axis of the format $(before_1, after_1, ... , before_N, after_N)$. It should be of length $2*N$ where $N$ is the number of dimensions of the array.This is equivalent to pad_width in numpy.pad, but flattened.
constant_value::double, optional, default=0: The value used for padding when mode is "constant".
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.pick — Method.

pick(data, index, axis, keepdims)

Picks elements from an input array according to the input indices along the given axis.

Given an input array of shape $(d0, d1)$ and indices of shape $(i0,)$, the result will be an output array of shape $(i0,)$ with::

output[i] = input[i, indices[i]]

By default, if any index mentioned is too large, it is replaced by the index that addresses the last element along an axis (the clip mode).

This function supports n-dimensional input and (n-1)-dimensional indices arrays.

Examples::

x = [[ 1., 2.], [ 3., 4.], [ 5., 6.]]

// picks elements with specified indices along axis 0 pick(x, y=[0,1], 0) = [ 1., 4.]

// picks elements with specified indices along axis 1 pick(x, y=[0,1,0], 1) = [ 1., 4., 5.]

y = [[ 1.], [ 0.], [ 2.]]

// picks elements with specified indices along axis 1 and dims are maintained pick(x,y, 1, keepdims=True) = [[ 2.], [ 3.], [ 6.]]

Defined in src/operator/tensor/broadcast_reduce_op_index.cc:L145

Arguments

data::NDArray-or-SymbolicNode: The input array
index::NDArray-or-SymbolicNode: The index array
axis::int or None, optional, default='None': The axis along which to perform the reduction. Negative values means indexing from right to left. $Requires axis to be set as int, because global reduction is not supported yet.$
keepdims::boolean, optional, default=0: If this is set to True, the reduced axis is left in the result as dimension with size one.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.radians — Method.

radians(data)

Converts each element of the input array from degrees to radians.

.. math:: radians([0, 90, 180, 270, 360]) = [0, \pi/2, \pi, 3\pi/2, 2\pi]

The storage type of $radians$ output depends upon the input storage type:

radians(default) = default
radians(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_trig.cc:L182

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.random_exponential — Method.

random_exponential(lam, shape, ctx, dtype)

random_exponential is an alias of _random_exponential.

Draw random samples from an exponential distribution.

Samples are distributed according to an exponential distribution parametrized by lambda (rate).

Example::

exponential(lam=4, shape=(2,2)) = [[ 0.0097189 , 0.08999364], [ 0.04146638, 0.31715935]]

Defined in src/operator/random/sample_op.cc:L115

Arguments

lam::float, optional, default=1: Lambda parameter (rate) of the exponential distribution.
shape::Shape(tuple), optional, default=[]: Shape of the output.
ctx::string, optional, default='': Context of output, in format cpu|gpu|cpu_pinned. Only used for imperative calls.
dtype::{'None', 'float16', 'float32', 'float64'},optional, default='None': DType of the output in case this can't be inferred. Defaults to float32 if not defined (dtype=None).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.random_gamma — Method.

random_gamma(alpha, beta, shape, ctx, dtype)

random_gamma is an alias of _random_gamma.

Draw random samples from a gamma distribution.

Samples are distributed according to a gamma distribution parametrized by alpha (shape) and beta (scale).

Example::

gamma(alpha=9, beta=0.5, shape=(2,2)) = [[ 7.10486984, 3.37695289], [ 3.91697288, 3.65933681]]

Defined in src/operator/random/sample_op.cc:L100

Arguments

alpha::float, optional, default=1: Alpha parameter (shape) of the gamma distribution.
beta::float, optional, default=1: Beta parameter (scale) of the gamma distribution.
shape::Shape(tuple), optional, default=[]: Shape of the output.
ctx::string, optional, default='': Context of output, in format cpu|gpu|cpu_pinned. Only used for imperative calls.
dtype::{'None', 'float16', 'float32', 'float64'},optional, default='None': DType of the output in case this can't be inferred. Defaults to float32 if not defined (dtype=None).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.random_generalized_negative_binomial — Method.

random_generalized_negative_binomial(mu, alpha, shape, ctx, dtype)

random_generalized_negative_binomial is an alias of _random_generalized_negative_binomial.

Draw random samples from a generalized negative binomial distribution.

Samples are distributed according to a generalized negative binomial distribution parametrized by mu (mean) and alpha (dispersion). alpha is defined as 1/k where k is the failure limit of the number of unsuccessful experiments (generalized to real numbers). Samples will always be returned as a floating point data type.

Example::

generalized_negative_binomial(mu=2.0, alpha=0.3, shape=(2,2)) = [[ 2., 1.], [ 6., 4.]]

Defined in src/operator/random/sample_op.cc:L168

Arguments

mu::float, optional, default=1: Mean of the negative binomial distribution.
alpha::float, optional, default=1: Alpha (dispersion) parameter of the negative binomial distribution.
shape::Shape(tuple), optional, default=[]: Shape of the output.
ctx::string, optional, default='': Context of output, in format cpu|gpu|cpu_pinned. Only used for imperative calls.
dtype::{'None', 'float16', 'float32', 'float64'},optional, default='None': DType of the output in case this can't be inferred. Defaults to float32 if not defined (dtype=None).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.random_negative_binomial — Method.

random_negative_binomial(k, p, shape, ctx, dtype)

random_negative_binomial is an alias of _random_negative_binomial.

Draw random samples from a negative binomial distribution.

Samples are distributed according to a negative binomial distribution parametrized by k (limit of unsuccessful experiments) and p (failure probability in each experiment). Samples will always be returned as a floating point data type.

Example::

negative_binomial(k=3, p=0.4, shape=(2,2)) = [[ 4., 7.], [ 2., 5.]]

Defined in src/operator/random/sample_op.cc:L149

Arguments

k::int, optional, default='1': Limit of unsuccessful experiments.
p::float, optional, default=1: Failure probability in each experiment.
shape::Shape(tuple), optional, default=[]: Shape of the output.
ctx::string, optional, default='': Context of output, in format cpu|gpu|cpu_pinned. Only used for imperative calls.
dtype::{'None', 'float16', 'float32', 'float64'},optional, default='None': DType of the output in case this can't be inferred. Defaults to float32 if not defined (dtype=None).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.random_normal — Method.

random_normal(loc, scale, shape, ctx, dtype)

random_normal is an alias of _random_normal.

Draw random samples from a normal (Gaussian) distribution.

.. note:: The existing alias $normal$ is deprecated.

Samples are distributed according to a normal distribution parametrized by loc (mean) and scale (standard deviation).

Example::

normal(loc=0, scale=1, shape=(2,2)) = [[ 1.89171135, -1.16881478], [-1.23474145, 1.55807114]]

Defined in src/operator/random/sample_op.cc:L85

Arguments

loc::float, optional, default=0: Mean of the distribution.
scale::float, optional, default=1: Standard deviation of the distribution.
shape::Shape(tuple), optional, default=[]: Shape of the output.
ctx::string, optional, default='': Context of output, in format cpu|gpu|cpu_pinned. Only used for imperative calls.
dtype::{'None', 'float16', 'float32', 'float64'},optional, default='None': DType of the output in case this can't be inferred. Defaults to float32 if not defined (dtype=None).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.random_poisson — Method.

random_poisson(lam, shape, ctx, dtype)

random_poisson is an alias of _random_poisson.

Draw random samples from a Poisson distribution.

Samples are distributed according to a Poisson distribution parametrized by lambda (rate). Samples will always be returned as a floating point data type.

Example::

poisson(lam=4, shape=(2,2)) = [[ 5., 2.], [ 4., 6.]]

Defined in src/operator/random/sample_op.cc:L132

Arguments

lam::float, optional, default=1: Lambda parameter (rate) of the Poisson distribution.
shape::Shape(tuple), optional, default=[]: Shape of the output.
ctx::string, optional, default='': Context of output, in format cpu|gpu|cpu_pinned. Only used for imperative calls.
dtype::{'None', 'float16', 'float32', 'float64'},optional, default='None': DType of the output in case this can't be inferred. Defaults to float32 if not defined (dtype=None).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.random_uniform — Method.

random_uniform(low, high, shape, ctx, dtype)

random_uniform is an alias of _random_uniform.

Draw random samples from a uniform distribution.

.. note:: The existing alias $uniform$ is deprecated.

Samples are uniformly distributed over the half-open interval [low, high) (includes low, but excludes high).

Example::

uniform(low=0, high=1, shape=(2,2)) = [[ 0.60276335, 0.85794562], [ 0.54488319, 0.84725171]]

Defined in src/operator/random/sample_op.cc:L66

Arguments

low::float, optional, default=0: Lower bound of the distribution.
high::float, optional, default=1: Upper bound of the distribution.
shape::Shape(tuple), optional, default=[]: Shape of the output.
ctx::string, optional, default='': Context of output, in format cpu|gpu|cpu_pinned. Only used for imperative calls.
dtype::{'None', 'float16', 'float32', 'float64'},optional, default='None': DType of the output in case this can't be inferred. Defaults to float32 if not defined (dtype=None).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.rcbrt — Method.

rcbrt(data)

Returns element-wise inverse cube-root value of the input.

.. math:: rcbrt(x) = 1/\sqrt[3]{x}

Example::

rcbrt([1,8,-125]) = [1.0, 0.5, -0.2]

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L618

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.reciprocal — Method.

reciprocal(data)

Returns the reciprocal of the argument, element-wise.

Calculates 1/x.

Example::

reciprocal([-2, 1, 3, 1.6, 0.2]) = [-0.5, 1.0, 0.33333334, 0.625, 5.0]

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L363

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.reshape_like — Method.

reshape_like(lhs, rhs)

Reshape lhs to have the same shape as rhs.

Arguments

lhs::NDArray-or-SymbolicNode: First input.
rhs::NDArray-or-SymbolicNode: Second input.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.rint — Method.

rint(data)

Returns element-wise rounded value to the nearest integer of the input.

.. note::

For input $n.5$ $rint$ returns $n$ while $round$ returns $n+1$.
For input $-n.5$ both $rint$ and $round$ returns $-n-1$.

Example::

rint([-1.5, 1.5, -1.9, 1.9, 2.1]) = [-2., 1., -2., 2., 2.]

The storage type of $rint$ output depends upon the input storage type:

rint(default) = default
rint(row_sparse) = row_sparse

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L444

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.rmsprop_update — Method.

rmsprop_update(weight, grad, n, lr, gamma1, epsilon, wd, rescale_grad, clip_gradient, clip_weights)

Update function for RMSProp optimizer.

RMSprop is a variant of stochastic gradient descent where the gradients are divided by a cache which grows with the sum of squares of recent gradients?

RMSProp is similar to AdaGrad, a popular variant of SGD which adaptively tunes the learning rate of each parameter. AdaGrad lowers the learning rate for each parameter monotonically over the course of training. While this is analytically motivated for convex optimizations, it may not be ideal for non-convex problems. RMSProp deals with this heuristically by allowing the learning rates to rebound as the denominator decays over time.

Define the Root Mean Square (RMS) error criterion of the gradient as :math:RMS[g]_t = \sqrt{E[g^2]_t + \epsilon}, where :math:g represents gradient and :math:E[g^2]_t is the decaying average over past squared gradient.

The :math:E[g^2]_t is given by:

.. math:: E[g^2]t = \gamma * E[g^2] + (1-\gamma) * g_t^2

The update step is

.. math:: \theta_{t+1} = \theta_t - \frac{\eta}{RMS[g]_t} g_t

The RMSProp code follows the version in http://www.cs.toronto.edu/~tijmen/csc321/slides/lecture_slides_lec6.pdf Tieleman & Hinton, 2012.

Hinton suggests the momentum term :math:\gamma to be 0.9 and the learning rate :math:\eta to be 0.001.

Defined in src/operator/optimizer_op.cc:L441

Arguments

weight::NDArray-or-SymbolicNode: Weight
grad::NDArray-or-SymbolicNode: Gradient
n::NDArray-or-SymbolicNode: n
lr::float, required: Learning rate
gamma1::float, optional, default=0.95: The decay rate of momentum estimates.
epsilon::float, optional, default=1e-08: A small constant for numerical stability.
wd::float, optional, default=0: Weight decay augments the objective function with a regularization term that penalizes large weights. The penalty scales with the square of the magnitude of each weight.
rescale_grad::float, optional, default=1: Rescale gradient to grad = rescale_grad*grad.
clip_gradient::float, optional, default=-1: Clip gradient to the range of [-clip_gradient, clip_gradient] If clip_gradient <= 0, gradient clipping is turned off. grad = max(min(grad, clip_gradient), -clip_gradient).
clip_weights::float, optional, default=-1: Clip weights to the range of [-clip_weights, clip_weights] If clip_weights <= 0, weight clipping is turned off. weights = max(min(weights, clip_weights), -clip_weights).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.rmspropalex_update — Method.

rmspropalex_update(weight, grad, n, g, delta, lr, gamma1, gamma2, epsilon, wd, rescale_grad, clip_gradient, clip_weights)

Update function for RMSPropAlex optimizer.

RMSPropAlex is non-centered version of RMSProp.

Define :math:E[g^2]_t is the decaying average over past squared gradient and :math:E[g]_t is the decaying average over past gradient.

.. math:: E[g^2]t = \gamma_1 * E[g^2] + (1 - \gamma_1) * g_t^2\ E[g]t = \gamma_1 * E[g] + (1 - \gamma_1) * g_t\ \Delta_t = \gamma_2 * \Delta_{t-1} - \frac{\eta}{\sqrt{E[g^2]_t - E[g]_t^2 + \epsilon}} g_t\ The update step is

.. math:: \theta_{t+1} = \theta_t + \Delta_t

The RMSPropAlex code follows the version in http://arxiv.org/pdf/1308.0850v5.pdf Eq(38) - Eq(45) by Alex Graves, 2013.

Graves suggests the momentum term :math:\gamma_1 to be 0.95, :math:\gamma_2 to be 0.9 and the learning rate :math:\eta to be 0.0001.

Defined in src/operator/optimizer_op.cc:L480

Arguments

weight::NDArray-or-SymbolicNode: Weight
grad::NDArray-or-SymbolicNode: Gradient
n::NDArray-or-SymbolicNode: n
g::NDArray-or-SymbolicNode: g
delta::NDArray-or-SymbolicNode: delta
lr::float, required: Learning rate
gamma1::float, optional, default=0.95: Decay rate.
gamma2::float, optional, default=0.9: Decay rate.
epsilon::float, optional, default=1e-08: A small constant for numerical stability.
wd::float, optional, default=0: Weight decay augments the objective function with a regularization term that penalizes large weights. The penalty scales with the square of the magnitude of each weight.
rescale_grad::float, optional, default=1: Rescale gradient to grad = rescale_grad*grad.
clip_gradient::float, optional, default=-1: Clip gradient to the range of [-clip_gradient, clip_gradient] If clip_gradient <= 0, gradient clipping is turned off. grad = max(min(grad, clip_gradient), -clip_gradient).
clip_weights::float, optional, default=-1: Clip weights to the range of [-clip_weights, clip_weights] If clip_weights <= 0, weight clipping is turned off. weights = max(min(weights, clip_weights), -clip_weights).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.rsqrt — Method.

rsqrt(data)

Returns element-wise inverse square-root value of the input.

.. math:: rsqrt(x) = 1/\sqrt{x}

Example::

rsqrt([4,9,16]) = [0.5, 0.33333334, 0.25]

The storage type of $rsqrt$ output is always dense

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L584

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.sample_exponential — Method.

sample_exponential(lam, shape, dtype)

sample_exponential is an alias of _sample_exponential.

Concurrent sampling from multiple exponential distributions with parameters lambda (rate).

The parameters of the distributions are provided as an input array. Let [s] be the shape of the input array, n be the dimension of [s], [t] be the shape specified as the parameter of the operator, and m be the dimension of [t]. Then the output will be a (n+m)-dimensional array with shape [s]x[t].

For any valid n-dimensional index i with respect to the input array, output[i] will be an m-dimensional array that holds randomly drawn samples from the distribution which is parameterized by the input value at index i. If the shape parameter of the operator is not set, then one sample will be drawn per distribution and the output array has the same shape as the input array.

Examples::

lam = [ 1.0, 8.5 ]

// Draw a single sample for each distribution sample_exponential(lam) = [ 0.51837951, 0.09994757]

// Draw a vector containing two samples for each distribution sample_exponential(lam, shape=(2)) = [[ 0.51837951, 0.19866663], [ 0.09994757, 0.50447971]]

Defined in src/operator/random/multisample_op.cc:L284

Arguments

lam::NDArray-or-SymbolicNode: Lambda (rate) parameters of the distributions.
shape::Shape(tuple), optional, default=[]: Shape to be sampled from each random distribution.
dtype::{'None', 'float16', 'float32', 'float64'},optional, default='None': DType of the output in case this can't be inferred. Defaults to float32 if not defined (dtype=None).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.sample_gamma — Method.

sample_gamma(alpha, shape, dtype, beta)

sample_gamma is an alias of _sample_gamma.

Concurrent sampling from multiple gamma distributions with parameters alpha (shape) and beta (scale).

The parameters of the distributions are provided as input arrays. Let [s] be the shape of the input arrays, n be the dimension of [s], [t] be the shape specified as the parameter of the operator, and m be the dimension of [t]. Then the output will be a (n+m)-dimensional array with shape [s]x[t].

For any valid n-dimensional index i with respect to the input arrays, output[i] will be an m-dimensional array that holds randomly drawn samples from the distribution which is parameterized by the input values at index i. If the shape parameter of the operator is not set, then one sample will be drawn per distribution and the output array has the same shape as the input arrays.

Examples::

alpha = [ 0.0, 2.5 ] beta = [ 1.0, 0.7 ]

// Draw a single sample for each distribution sample_gamma(alpha, beta) = [ 0. , 2.25797319]

// Draw a vector containing two samples for each distribution sample_gamma(alpha, beta, shape=(2)) = [[ 0. , 0. ], [ 2.25797319, 1.70734084]]

Defined in src/operator/random/multisample_op.cc:L282

Arguments

alpha::NDArray-or-SymbolicNode: Alpha (shape) parameters of the distributions.
shape::Shape(tuple), optional, default=[]: Shape to be sampled from each random distribution.
dtype::{'None', 'float16', 'float32', 'float64'},optional, default='None': DType of the output in case this can't be inferred. Defaults to float32 if not defined (dtype=None).
beta::NDArray-or-SymbolicNode: Beta (scale) parameters of the distributions.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.sample_generalized_negative_binomial — Method.

sample_generalized_negative_binomial(mu, shape, dtype, alpha)

sample_generalized_negative_binomial is an alias of _sample_generalized_negative_binomial.

Concurrent sampling from multiple generalized negative binomial distributions with parameters mu (mean) and alpha (dispersion).

The parameters of the distributions are provided as input arrays. Let [s] be the shape of the input arrays, n be the dimension of [s], [t] be the shape specified as the parameter of the operator, and m be the dimension of [t]. Then the output will be a (n+m)-dimensional array with shape [s]x[t].

For any valid n-dimensional index i with respect to the input arrays, output[i] will be an m-dimensional array that holds randomly drawn samples from the distribution which is parameterized by the input values at index i. If the shape parameter of the operator is not set, then one sample will be drawn per distribution and the output array has the same shape as the input arrays.

Samples will always be returned as a floating point data type.

Examples::

mu = [ 2.0, 2.5 ] alpha = [ 1.0, 0.1 ]

// Draw a single sample for each distribution sample_generalized_negative_binomial(mu, alpha) = [ 0., 3.]

// Draw a vector containing two samples for each distribution sample_generalized_negative_binomial(mu, alpha, shape=(2)) = [[ 0., 3.], [ 3., 1.]]

Defined in src/operator/random/multisample_op.cc:L293

Arguments

mu::NDArray-or-SymbolicNode: Means of the distributions.
shape::Shape(tuple), optional, default=[]: Shape to be sampled from each random distribution.
dtype::{'None', 'float16', 'float32', 'float64'},optional, default='None': DType of the output in case this can't be inferred. Defaults to float32 if not defined (dtype=None).
alpha::NDArray-or-SymbolicNode: Alpha (dispersion) parameters of the distributions.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.sample_multinomial — Method.

sample_multinomial(data, shape, get_prob, dtype)

sample_multinomial is an alias of _sample_multinomial.

Concurrent sampling from multiple multinomial distributions.

data is an n dimensional array whose last dimension has length k, where k is the number of possible outcomes of each multinomial distribution. This operator will draw shape samples from each distribution. If shape is empty one sample will be drawn from each distribution.

If get_prob is true, a second array containing log likelihood of the drawn samples will also be returned. This is usually used for reinforcement learning where you can provide reward as head gradient for this array to estimate gradient.

Note that the input distribution must be normalized, i.e. data must sum to 1 along its last axis.

Examples::

probs = [[0, 0.1, 0.2, 0.3, 0.4], [0.4, 0.3, 0.2, 0.1, 0]]

// Draw a single sample for each distribution sample_multinomial(probs) = [3, 0]

// Draw a vector containing two samples for each distribution sample_multinomial(probs, shape=(2)) = [[4, 2], [0, 0]]

// requests log likelihood sample_multinomial(probs, get_prob=True) = [2, 1], [0.2, 0.3]

Arguments

data::NDArray-or-SymbolicNode: Distribution probabilities. Must sum to one on the last axis.
shape::Shape(tuple), optional, default=[]: Shape to be sampled from each random distribution.
get_prob::boolean, optional, default=0: Whether to also return the log probability of sampled result. This is usually used for differentiating through stochastic variables, e.g. in reinforcement learning.
dtype::{'int32'},optional, default='int32': DType of the output in case this can't be inferred. Only support int32 for now.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.sample_negative_binomial — Method.

sample_negative_binomial(k, shape, dtype, p)

sample_negative_binomial is an alias of _sample_negative_binomial.

Concurrent sampling from multiple negative binomial distributions with parameters k (failure limit) and p (failure probability).

The parameters of the distributions are provided as input arrays. Let [s] be the shape of the input arrays, n be the dimension of [s], [t] be the shape specified as the parameter of the operator, and m be the dimension of [t]. Then the output will be a (n+m)-dimensional array with shape [s]x[t].

For any valid n-dimensional index i with respect to the input arrays, output[i] will be an m-dimensional array that holds randomly drawn samples from the distribution which is parameterized by the input values at index i. If the shape parameter of the operator is not set, then one sample will be drawn per distribution and the output array has the same shape as the input arrays.

Samples will always be returned as a floating point data type.

Examples::

k = [ 20, 49 ] p = [ 0.4 , 0.77 ]

// Draw a single sample for each distribution sample_negative_binomial(k, p) = [ 15., 16.]

// Draw a vector containing two samples for each distribution sample_negative_binomial(k, p, shape=(2)) = [[ 15., 50.], [ 16., 12.]]

Defined in src/operator/random/multisample_op.cc:L289

Arguments

k::NDArray-or-SymbolicNode: Limits of unsuccessful experiments.
shape::Shape(tuple), optional, default=[]: Shape to be sampled from each random distribution.
dtype::{'None', 'float16', 'float32', 'float64'},optional, default='None': DType of the output in case this can't be inferred. Defaults to float32 if not defined (dtype=None).
p::NDArray-or-SymbolicNode: Failure probabilities in each experiment.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.sample_normal — Method.

sample_normal(mu, shape, dtype, sigma)

sample_normal is an alias of _sample_normal.

Concurrent sampling from multiple normal distributions with parameters mu (mean) and sigma (standard deviation).

The parameters of the distributions are provided as input arrays. Let [s] be the shape of the input arrays, n be the dimension of [s], [t] be the shape specified as the parameter of the operator, and m be the dimension of [t]. Then the output will be a (n+m)-dimensional array with shape [s]x[t].

For any valid n-dimensional index i with respect to the input arrays, output[i] will be an m-dimensional array that holds randomly drawn samples from the distribution which is parameterized by the input values at index i. If the shape parameter of the operator is not set, then one sample will be drawn per distribution and the output array has the same shape as the input arrays.

Examples::

mu = [ 0.0, 2.5 ] sigma = [ 1.0, 3.7 ]

// Draw a single sample for each distribution sample_normal(mu, sigma) = [-0.56410581, 0.95934606]

// Draw a vector containing two samples for each distribution sample_normal(mu, sigma, shape=(2)) = [[-0.56410581, 0.2928229 ], [ 0.95934606, 4.48287058]]

Defined in src/operator/random/multisample_op.cc:L279

Arguments

mu::NDArray-or-SymbolicNode: Means of the distributions.
shape::Shape(tuple), optional, default=[]: Shape to be sampled from each random distribution.
dtype::{'None', 'float16', 'float32', 'float64'},optional, default='None': DType of the output in case this can't be inferred. Defaults to float32 if not defined (dtype=None).
sigma::NDArray-or-SymbolicNode: Standard deviations of the distributions.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.sample_poisson — Method.

sample_poisson(lam, shape, dtype)

sample_poisson is an alias of _sample_poisson.

Concurrent sampling from multiple Poisson distributions with parameters lambda (rate).

The parameters of the distributions are provided as an input array. Let [s] be the shape of the input array, n be the dimension of [s], [t] be the shape specified as the parameter of the operator, and m be the dimension of [t]. Then the output will be a (n+m)-dimensional array with shape [s]x[t].

For any valid n-dimensional index i with respect to the input array, output[i] will be an m-dimensional array that holds randomly drawn samples from the distribution which is parameterized by the input value at index i. If the shape parameter of the operator is not set, then one sample will be drawn per distribution and the output array has the same shape as the input array.

Samples will always be returned as a floating point data type.

Examples::

lam = [ 1.0, 8.5 ]

// Draw a single sample for each distribution sample_poisson(lam) = [ 0., 13.]

// Draw a vector containing two samples for each distribution sample_poisson(lam, shape=(2)) = [[ 0., 4.], [ 13., 8.]]

Defined in src/operator/random/multisample_op.cc:L286

Arguments

lam::NDArray-or-SymbolicNode: Lambda (rate) parameters of the distributions.
shape::Shape(tuple), optional, default=[]: Shape to be sampled from each random distribution.
dtype::{'None', 'float16', 'float32', 'float64'},optional, default='None': DType of the output in case this can't be inferred. Defaults to float32 if not defined (dtype=None).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.sample_uniform — Method.

sample_uniform(low, shape, dtype, high)

sample_uniform is an alias of _sample_uniform.

Concurrent sampling from multiple uniform distributions on the intervals given by [low,high).

The parameters of the distributions are provided as input arrays. Let [s] be the shape of the input arrays, n be the dimension of [s], [t] be the shape specified as the parameter of the operator, and m be the dimension of [t]. Then the output will be a (n+m)-dimensional array with shape [s]x[t].

For any valid n-dimensional index i with respect to the input arrays, output[i] will be an m-dimensional array that holds randomly drawn samples from the distribution which is parameterized by the input values at index i. If the shape parameter of the operator is not set, then one sample will be drawn per distribution and the output array has the same shape as the input arrays.

Examples::

low = [ 0.0, 2.5 ] high = [ 1.0, 3.7 ]

// Draw a single sample for each distribution sample_uniform(low, high) = [ 0.40451524, 3.18687344]

// Draw a vector containing two samples for each distribution sample_uniform(low, high, shape=(2)) = [[ 0.40451524, 0.18017688], [ 3.18687344, 3.68352246]]

Defined in src/operator/random/multisample_op.cc:L277

Arguments

low::NDArray-or-SymbolicNode: Lower bounds of the distributions.
shape::Shape(tuple), optional, default=[]: Shape to be sampled from each random distribution.
dtype::{'None', 'float16', 'float32', 'float64'},optional, default='None': DType of the output in case this can't be inferred. Defaults to float32 if not defined (dtype=None).
high::NDArray-or-SymbolicNode: Upper bounds of the distributions.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.save — Method.

save(filename :: AbstractString, node :: SymbolicNode)

Save a SymbolicNode to a JSON file.

source

# MXNet.mx.scatter_nd — Method.

scatter_nd(data, indices, shape)

Scatters data into a new tensor according to indices.

Given data with shape (Y_0, ..., Y_{K-1}, X_M, ..., X_{N-1}) and indices with shape (M, Y_0, ..., Y_{K-1}), the output will have shape (X_0, X_1, ..., X_{N-1}), where M <= N. If M == N, data shape should simply be (Y_0, ..., Y_{K-1}).

The elements in output is defined as follows::

output[indices[0, y_0, ..., y_{K-1}], ..., indices[M-1, y_0, ..., y_{K-1}], x_M, ..., x_{N-1}] = data[y_0, ..., y_{K-1}, x_M, ..., x_{N-1}]

all other entries in output are 0.

.. warning::

If the indices have duplicates, the result will be non-deterministic and
the gradient of `scatter_nd` will not be correct!!

Examples::

data = [2, 3, 0] indices = [[1, 1, 0], [0, 1, 0]] shape = (2, 2) scatter_nd(data, indices, shape) = [[0, 0], [2, 3]]

Arguments

data::NDArray-or-SymbolicNode: data
indices::NDArray-or-SymbolicNode: indices
shape::Shape(tuple), required: Shape of output.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.set_attr — Method.

set_attr(self:: SymbolicNode, key :: Symbol, value :: AbstractString)

Set the attribute key to value for this SymbolicNode.

Note

It is encouraged not to call this function directly, unless you know exactly what you are doing. The recommended way of setting attributes is when creating the SymbolicNode. Changing the attributes of a SymbolicNode that is already been used somewhere else might cause unexpected behavior and inconsistency.

source

# MXNet.mx.sgd_mom_update — Method.

sgd_mom_update(weight, grad, mom, lr, momentum, wd, rescale_grad, clip_gradient)

Momentum update function for Stochastic Gradient Descent (SDG) optimizer.

Momentum update has better convergence rates on neural networks. Mathematically it looks like below:

.. math::

v_1 = \alpha * \nabla J(W_0)\ v_t = \gamma v_{t-1} - \alpha * \nabla J(W_{t-1})\ W_t = W_{t-1} + v_t

It updates the weights using::

v = momentum * v - learning_rate * gradient weight += v

Where the parameter $momentum$ is the decay rate of momentum estimates at each epoch.

If weight and grad are both of $row_sparse$ storage type and momentum is of $default$ storage type, standard update is applied.

If weight, grad and momentum are all of $row_sparse$ storage type, only the row slices whose indices appear in grad.indices are updated (for both weight and momentum)::

for row in gradient.indices: v[row] = momentum[row] * v[row] - learning_rate * gradient[row] weight[row] += v[row]

Defined in src/operator/optimizer_op.cc:L265

Arguments

weight::NDArray-or-SymbolicNode: Weight
grad::NDArray-or-SymbolicNode: Gradient
mom::NDArray-or-SymbolicNode: Momentum
lr::float, required: Learning rate
momentum::float, optional, default=0: The decay rate of momentum estimates at each epoch.
wd::float, optional, default=0: Weight decay augments the objective function with a regularization term that penalizes large weights. The penalty scales with the square of the magnitude of each weight.
rescale_grad::float, optional, default=1: Rescale gradient to grad = rescale_grad*grad.
clip_gradient::float, optional, default=-1: Clip gradient to the range of [-clip_gradient, clip_gradient] If clip_gradient <= 0, gradient clipping is turned off. grad = max(min(grad, clip_gradient), -clip_gradient).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.sgd_update — Method.

sgd_update(weight, grad, lr, wd, rescale_grad, clip_gradient)

Update function for Stochastic Gradient Descent (SDG) optimizer.

It updates the weights using::

weight = weight - learning_rate * gradient

If weight is of $row_sparse$ storage type, only the row slices whose indices appear in grad.indices are updated::

for row in gradient.indices: weight[row] = weight[row] - learning_rate * gradient[row]

Defined in src/operator/optimizer_op.cc:L222

Arguments

weight::NDArray-or-SymbolicNode: Weight
grad::NDArray-or-SymbolicNode: Gradient
lr::float, required: Learning rate
wd::float, optional, default=0: Weight decay augments the objective function with a regularization term that penalizes large weights. The penalty scales with the square of the magnitude of each weight.
rescale_grad::float, optional, default=1: Rescale gradient to grad = rescale_grad*grad.
clip_gradient::float, optional, default=-1: Clip gradient to the range of [-clip_gradient, clip_gradient] If clip_gradient <= 0, gradient clipping is turned off. grad = max(min(grad, clip_gradient), -clip_gradient).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.signsgd_update — Method.

signsgd_update(weight, grad, lr, wd, rescale_grad, clip_gradient)

Update function for SignSGD optimizer. .. math::

g_t = \nabla J(W_{t-1})\ W_t = W_{t-1} - \eta_t \text{sign}(g_t)}

It updates the weights using::

weight = weight - learning_rate * sign(gradient)

.. note::

sparse ndarray not supported for this optimizer yet.

Defined in src/operator/optimizer_op.cc:L55

Arguments

weight::NDArray-or-SymbolicNode: Weight
grad::NDArray-or-SymbolicNode: Gradient
lr::float, required: Learning rate
wd::float, optional, default=0: Weight decay augments the objective function with a regularization term that penalizes large weights. The penalty scales with the square of the magnitude of each weight.
rescale_grad::float, optional, default=1: Rescale gradient to grad = rescale_grad*grad.
clip_gradient::float, optional, default=-1: Clip gradient to the range of [-clip_gradient, clip_gradient] If clip_gradient <= 0, gradient clipping is turned off. grad = max(min(grad, clip_gradient), -clip_gradient).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.signum_update — Method.

signum_update(weight, grad, mom, lr, momentum, wd, rescale_grad, clip_gradient, wd_lh)

SIGN momentUM (Signum) optimizer.

.. math::

g_t = \nabla J(W_{t-1})\ m_t = \beta m_{t-1} + (1 - \beta) g_t\ W_t = W_{t-1} - \eta_t \text{sign}(m_t)}

It updates the weights using:: state = momentum * state + (1-momentum) * gradient weight = weight - learning_rate * sign(state)

Where the parameter $momentum$ is the decay rate of momentum estimates at each epoch.

.. note::

sparse ndarray not supported for this optimizer yet.

Defined in src/operator/optimizer_op.cc:L84

Arguments

weight::NDArray-or-SymbolicNode: Weight
grad::NDArray-or-SymbolicNode: Gradient
mom::NDArray-or-SymbolicNode: Momentum
lr::float, required: Learning rate
momentum::float, optional, default=0: The decay rate of momentum estimates at each epoch.
wd::float, optional, default=0: Weight decay augments the objective function with a regularization term that penalizes large weights. The penalty scales with the square of the magnitude of each weight.
rescale_grad::float, optional, default=1: Rescale gradient to grad = rescale_grad*grad.
clip_gradient::float, optional, default=-1: Clip gradient to the range of [-clip_gradient, clip_gradient] If clip_gradient <= 0, gradient clipping is turned off. grad = max(min(grad, clip_gradient), -clip_gradient).
wd_lh::float, optional, default=0: The amount of weight decay that does not go into gradient/momentum calculationsotherwise do weight decay algorithmically only.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.slice — Method.

slice(data, begin, end, step)

Slices a region of the array.

.. note:: $crop$ is deprecated. Use $slice$ instead.

This function returns a sliced array between the indices given by begin and end with the corresponding step.

For an input array of $shape=(d_0, d_1, ..., d_n-1)$, slice operation with $begin=(b_0, b_1...b_m-1)$, $end=(e_0, e_1, ..., e_m-1)$, and $step=(s_0, s_1, ..., s_m-1)$, where m <= n, results in an array with the shape $(|e_0-b_0|/|s_0|, ..., |e_m-1-b_m-1|/|s_m-1|, d_m, ..., d_n-1)$.

The resulting array's k-th dimension contains elements from the k-th dimension of the input array starting from index $b_k$ (inclusive) with step $s_k$ until reaching $e_k$ (exclusive).

If the k-th elements are None in the sequence of begin, end, and step, the following rule will be used to set default values. If s_k is None, set s_k=1. If s_k > 0, set b_k=0, e_k=d_k; else, set b_k=d_k-1, e_k=-1.

The storage type of $slice$ output depends on storage types of inputs

slice(csr) = csr
otherwise, $slice$ generates output with default storage

.. note:: When input data storage type is csr, it only supports step=(), or step=(None,), or step=(1,) to generate a csr output. For other step parameter values, it falls back to slicing a dense tensor.

Example::

x = [[ 1., 2., 3., 4.], [ 5., 6., 7., 8.], [ 9., 10., 11., 12.]]

slice(x, begin=(0,1), end=(2,4)) = [[ 2., 3., 4.], [ 6., 7., 8.]] slice(x, begin=(None, 0), end=(None, 3), step=(-1, 2)) = [[9., 11.], [5., 7.], [1., 3.]]

Defined in src/operator/tensor/matrix_op.cc:L359

Arguments

data::NDArray-or-SymbolicNode: Source input
begin::Shape(tuple), required: starting indices for the slice operation, supports negative indices.
end::Shape(tuple), required: ending indices for the slice operation, supports negative indices.
step::Shape(tuple), optional, default=[]: step for the slice operation, supports negative values.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.slice_axis — Method.

slice_axis(data, axis, begin, end)

Slices along a given axis.

Returns an array slice along a given axis starting from the begin index to the end index.

Examples::

x = [[ 1., 2., 3., 4.], [ 5., 6., 7., 8.], [ 9., 10., 11., 12.]]

slice_axis(x, axis=0, begin=1, end=3) = [[ 5., 6., 7., 8.], [ 9., 10., 11., 12.]]

slice_axis(x, axis=1, begin=0, end=2) = [[ 1., 2.], [ 5., 6.], [ 9., 10.]]

slice_axis(x, axis=1, begin=-3, end=-1) = [[ 2., 3.], [ 6., 7.], [ 10., 11.]]

Defined in src/operator/tensor/matrix_op.cc:L446

Arguments

data::NDArray-or-SymbolicNode: Source input
axis::int, required: Axis along which to be sliced, supports negative indexes.
begin::int, required: The beginning index along the axis to be sliced, supports negative indexes.
end::int or None, required: The ending index along the axis to be sliced, supports negative indexes.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.smooth_l1 — Method.

smooth_l1(data, scalar)

Calculate Smooth L1 Loss(lhs, scalar) by summing

.. math::

f(x) =
\begin{cases}
(\sigma x)^2/2,& \text{if }x < 1/\sigma^2\\
|x|-0.5/\sigma^2,& \text{otherwise}
\end{cases}

where :math:x is an element of the tensor lhs and :math:\sigma is the scalar.

Example::

smooth_l1([1, 2, 3, 4], sigma=1) = [0.5, 1.5, 2.5, 3.5]

Defined in src/operator/tensor/elemwise_binary_scalar_op_extended.cc:L103

Arguments

data::NDArray-or-SymbolicNode: source input
scalar::float: scalar input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.softmax_cross_entropy — Method.

softmax_cross_entropy(data, label)

Calculate cross entropy of softmax output and one-hot label.

This operator computes the cross entropy in two steps:
- Applies softmax function on the input array.
- Computes and returns the cross entropy loss between the softmax output and the labels.
- The softmax function and cross entropy loss is given by:
- Softmax Function:
.. math:: \text{softmax}(x)_i = \frac{exp(x_i)}{\sum_j exp(x_j)}
- Cross Entropy Function:
.. math:: \text{CE(label, output)} = - \sum_i \text{label}_i \log(\text{output}_i)

Example::

x = [[1, 2, 3], [11, 7, 5]]

label = [2, 0]

softmax(x) = [[0.09003057, 0.24472848, 0.66524094], [0.97962922, 0.01794253, 0.00242826]]

softmax_cross_entropy(data, label) = - log(0.66524084) - log(0.97962922) = 0.4281871

Defined in src/operator/loss_binary_op.cc:L59

Arguments

data::NDArray-or-SymbolicNode: Input data
label::NDArray-or-SymbolicNode: Input label
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.square — Method.

square(data)

Returns element-wise squared value of the input.

.. math:: square(x) = x^2

Example::

square([2, 3, 4]) = [4, 9, 16]

The storage type of $square$ output depends upon the input storage type:

square(default) = default
square(row_sparse) = row_sparse
square(csr) = csr

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L541

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.stack — Method.

stack(data, axis, num_args)

Note: stack takes variable number of positional inputs. So instead of calling as stack([x, y, z], num_args=3), one should call via stack(x, y, z), and num_args will be determined automatically.

Join a sequence of arrays along a new axis.

The axis parameter specifies the index of the new axis in the dimensions of the result. For example, if axis=0 it will be the first dimension and if axis=-1 it will be the last dimension.

Examples::

x = [1, 2] y = [3, 4]

stack(x, y) = [[1, 2], [3, 4]] stack(x, y, axis=1) = [[1, 3], [2, 4]]

Arguments

data::NDArray-or-SymbolicNode[]: List of arrays to stack
axis::int, optional, default='0': The axis in the result array along which the input arrays are stacked.
num_args::int, required: Number of inputs to be stacked.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.stop_gradient — Method.

stop_gradient(data)

stop_gradient is an alias of BlockGrad.

Stops gradient computation.

Stops the accumulated gradient of the inputs from flowing through this operator in the backward direction. In other words, this operator prevents the contribution of its inputs to be taken into account for computing gradients.

Example::

v1 = [1, 2] v2 = [0, 1] a = Variable('a') b = Variable('b') b_stop_grad = stop_gradient(3 * b) loss = MakeLoss(b_stop_grad + a)

executor = loss.simple_bind(ctx=cpu(), a=(1,2), b=(1,2)) executor.forward(is_train=True, a=v1, b=v2) executor.outputs [ 1. 5.]

executor.backward() executor.grad_arrays [ 0. 0.] [ 1. 1.]

Defined in src/operator/tensor/elemwise_unary_op_basic.cc:L166

Arguments

data::NDArray-or-SymbolicNode: The input array.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.sum_axis — Method.

sum_axis(data, axis, keepdims, exclude)

sum_axis is an alias of sum.

Computes the sum of array elements over given axes.

.. Note::

sum and sum_axis are equivalent. For ndarray of csr storage type summation along axis 0 and axis 1 is supported. Setting keepdims or exclude to True will cause a fallback to dense operator.

Example::

data = [[[1,2],[2,3],[1,3]], [[1,4],[4,3],[5,2]], [[7,1],[7,2],[7,3]]]

sum(data, axis=1) [[ 4. 8.] [ 10. 9.] [ 21. 6.]]

sum(data, axis=[1,2]) [ 12. 19. 27.]

data = [[1,2,0], [3,0,1], [4,1,0]]

csr = cast_storage(data, 'csr')

sum(csr, axis=0) [ 8. 3. 1.]

sum(csr, axis=1) [ 3. 4. 5.]

Defined in src/operator/tensor/broadcast_reduce_op_value.cc:L85

Arguments

data::NDArray-or-SymbolicNode: The input
axis::Shape(tuple), optional, default=[]: The axis or axes along which to perform the reduction.

`` The default,axis=(), will compute over all elements into a scalar array with shape(1,)`.

If axis is int, a reduction is performed on a particular axis.

If axis is a tuple of ints, a reduction is performed on all the axes specified in the tuple.

If exclude is true, reduction will be performed on the axes that are NOT in axis instead.

Negative values means indexing from right to left. `` *keepdims::boolean, optional, default=0: If this is set toTrue, the reduced axes are left in the result as dimension with size one. *exclude::boolean, optional, default=0: Whether to perform reduction on axis that are NOT in axis instead. *name::Symbol: The name of theSymbolicNode. (e.g.:my_symbol), optional. *attrs::Dict{Symbol, AbstractString}: The attributes associated with thisSymbolicNode`.

source

# MXNet.mx.swapaxes — Method.

swapaxes(data, dim1, dim2)

swapaxes is an alias of SwapAxis.

Interchanges two axes of an array.

Examples::

x = [[1, 2, 3]]) swapaxes(x, 0, 1) = [[ 1], [ 2], [ 3]]

x = [[[ 0, 1], [ 2, 3]], [[ 4, 5], [ 6, 7]]] // (2,2,2) array

swapaxes(x, 0, 2) = [[[ 0, 4], [ 2, 6]], [[ 1, 5], [ 3, 7]]]

Defined in src/operator/swapaxis.cc:L70

Arguments

data::NDArray-or-SymbolicNode: Input array.
dim1::int (non-negative), optional, default=0: the first axis to be swapped.
dim2::int (non-negative), optional, default=0: the second axis to be swapped.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.take — Method.

take(a, indices, axis, mode)

Takes elements from an input array along the given axis.

This function slices the input array along a particular axis with the provided indices.

Given an input array with shape $(d0, d1, d2)$ and indices with shape $(i0, i1)$, the output will have shape $(i0, i1, d1, d2)$, computed by::

output[i,j,:,:] = input[indices[i,j],:,:]

.. note::

axis- Only slicing along axis 0 is supported for now.
mode- Only clip mode is supported for now.

Examples:: x = [4. 5. 6.]

// Trivial case, take the second element along the first axis. take(x, [1]) = [ 5. ]

x = [[ 1., 2.], [ 3., 4.], [ 5., 6.]]

// In this case we will get rows 0 and 1, then 1 and 2. Along axis 0 take(x, [[0,1],[1,2]]) = [[[ 1., 2.], [ 3., 4.]],

                        [[ 3.,  4.],
                         [ 5.,  6.]]]

Defined in src/operator/tensor/indexing_op.cc:L371

Arguments

a::NDArray-or-SymbolicNode: The input array.
indices::NDArray-or-SymbolicNode: The indices of the values to be extracted.
axis::int, optional, default='0': The axis of input array to be taken.
mode::{'clip', 'raise', 'wrap'},optional, default='clip': Specify how out-of-bound indices bahave. "clip" means clip to the range. So, if all indices mentioned are too large, they are replaced by the index that addresses the last element along an axis. "wrap" means to wrap around. "raise" means to raise an error.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.tile — Method.

tile(data, reps)

Repeats the whole array multiple times.

If $reps$ has length d, and input array has dimension of n. There are three cases:

n=d. Repeat i-th dimension of the input by $reps[i]$ times::

x = [[1, 2], [3, 4]]

tile(x, reps=(2,3)) = [[ 1., 2., 1., 2., 1., 2.], [ 3., 4., 3., 4., 3., 4.], [ 1., 2., 1., 2., 1., 2.], [ 3., 4., 3., 4., 3., 4.]] * n>d. $reps$ is promoted to length n by pre-pending 1's to it. Thus for an input shape $(2,3)$, $repos=(2,)$ is treated as $(1,2)$::

tile(x, reps=(2,)) = [[ 1.,  2.,  1.,  2.],
                      [ 3.,  4.,  3.,  4.]]

n<d. The input is promoted to be d-dimensional by prepending new axes. So a shape $(2,2)$ array is promoted to $(1,2,2)$ for 3-D replication::

tile(x, reps=(2,2,3)) = [[[ 1., 2., 1., 2., 1., 2.], [ 3., 4., 3., 4., 3., 4.], [ 1., 2., 1., 2., 1., 2.], [ 3., 4., 3., 4., 3., 4.]],

[[ 1., 2., 1., 2., 1., 2.], [ 3., 4., 3., 4., 3., 4.], [ 1., 2., 1., 2., 1., 2.], [ 3., 4., 3., 4., 3., 4.]]]

Defined in src/operator/tensor/matrix_op.cc:L624

Arguments

data::NDArray-or-SymbolicNode: Input data array
reps::Shape(tuple), required: The number of times for repeating the tensor a. If reps has length d, the result will have dimension of max(d, a.ndim); If a.ndim < d, a is promoted to be d-dimensional by prepending new axes. If a.ndim > d, reps is promoted to a.ndim by pre-pending 1's to it.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.to_json — Method.

to_json(self :: SymbolicNode)

Convert a SymbolicNode into a JSON string.

source

# MXNet.mx.topk — Method.

topk(data, axis, k, ret_typ, is_ascend)

Returns the top k elements in an input array along the given axis.

Examples::

x = [[ 0.3, 0.2, 0.4], [ 0.1, 0.3, 0.2]]

// returns an index of the largest element on last axis topk(x) = [[ 2.], [ 1.]]

// returns the value of top-2 largest elements on last axis topk(x, ret_typ='value', k=2) = [[ 0.4, 0.3], [ 0.3, 0.2]]

// returns the value of top-2 smallest elements on last axis topk(x, ret_typ='value', k=2, is_ascend=1) = [[ 0.2 , 0.3], [ 0.1 , 0.2]]

// returns the value of top-2 largest elements on axis 0 topk(x, axis=0, ret_typ='value', k=2) = [[ 0.3, 0.3, 0.4], [ 0.1, 0.2, 0.2]]

// flattens and then returns list of both values and indices topk(x, ret_typ='both', k=2) = [[[ 0.4, 0.3], [ 0.3, 0.2]] , [[ 2., 0.], [ 1., 2.]]]

Defined in src/operator/tensor/ordering_op.cc:L63

Arguments

data::NDArray-or-SymbolicNode: The input array
axis::int or None, optional, default='-1': Axis along which to choose the top k indices. If not given, the flattened array is used. Default is -1.
k::int, optional, default='1': Number of top elements to select, should be always smaller than or equal to the element number in the given axis. A global sort is performed if set k < 1.
ret_typ::{'both', 'indices', 'mask', 'value'},optional, default='indices': The return type.

"value" means to return the top k values, "indices" means to return the indices of the top k values, "mask" means to return a mask array containing 0 and 1. 1 means the top k values. "both" means to return a list of both values and indices of top k elements.

is_ascend::boolean, optional, default=0: Whether to choose k largest or k smallest elements. Top K largest elements will be chosen if set to false.
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.uniform — Method.

uniform(low, high, shape, ctx, dtype)

uniform is an alias of _random_uniform.

Draw random samples from a uniform distribution.

.. note:: The existing alias $uniform$ is deprecated.

Samples are uniformly distributed over the half-open interval [low, high) (includes low, but excludes high).

Example::

uniform(low=0, high=1, shape=(2,2)) = [[ 0.60276335, 0.85794562], [ 0.54488319, 0.84725171]]

Defined in src/operator/random/sample_op.cc:L66

Arguments

low::float, optional, default=0: Lower bound of the distribution.
high::float, optional, default=1: Upper bound of the distribution.
shape::Shape(tuple), optional, default=[]: Shape of the output.
ctx::string, optional, default='': Context of output, in format cpu|gpu|cpu_pinned. Only used for imperative calls.
dtype::{'None', 'float16', 'float32', 'float64'},optional, default='None': DType of the output in case this can't be inferred. Defaults to float32 if not defined (dtype=None).
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.where — Method.

where(condition, x, y)

Return the elements, either from x or y, depending on the condition.

Given three ndarrays, condition, x, and y, return an ndarray with the elements from x or y, depending on the elements from condition are true or false. x and y must have the same shape. If condition has the same shape as x, each element in the output array is from x if the corresponding element in the condition is true, and from y if false.

If condition does not have the same shape as x, it must be a 1D array whose size is the same as x's first dimension size. Each row of the output array is from x's row if the corresponding element from condition is true, and from y's row if false.

Note that all non-zero values are interpreted as $True$ in condition.

Examples::

x = [[1, 2], [3, 4]] y = [[5, 6], [7, 8]] cond = [[0, 1], [-1, 0]]

where(cond, x, y) = [[5, 2], [3, 8]]

csr_cond = cast_storage(cond, 'csr')

where(csr_cond, x, y) = [[5, 2], [3, 8]]

Defined in src/operator/tensor/control_flow_op.cc:L57

Arguments

condition::NDArray-or-SymbolicNode: condition array
x::NDArray-or-SymbolicNode:
y::NDArray-or-SymbolicNode:
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

# MXNet.mx.zeros_like — Method.

zeros_like(data)

Return an array of zeros with the same shape and type as the input array.

The storage type of $zeros_like$ output depends on the storage type of the input

zeros_like(row_sparse) = row_sparse
zeros_like(csr) = csr
zeros_like(default) = default

Examples::

x = [[ 1., 1., 1.], [ 1., 1., 1.]]

zeros_like(x) = [[ 0., 0., 0.], [ 0., 0., 0.]]

Arguments

data::NDArray-or-SymbolicNode: The input
name::Symbol: The name of the SymbolicNode. (e.g. :my_symbol), optional.
attrs::Dict{Symbol, AbstractString}: The attributes associated with this SymbolicNode.

source

Symbolic API

forward softmax output

backward gradient output

notice that the first row is all 0 because label[0] is 1, which is equal to ignore_label.