Base Operations

These ops are generally available directly on SameDiff instances. Due to an oversight before the release, this ops aren't also available on Nd4j. To use the INDArray variants of these operations, you will have to instantiate a NDBase instance.

all

INDArray all(INDArray x, int[] dimensions)

SDVariable all(SDVariable x, int[] dimensions)
SDVariable all(String name, SDVariable x, int[] dimensions)

Boolean and array reduction operation, optionally along specified dimensions

  • x (NDARRAY) - Input variable

  • dimensions - Dimensions to reduce over. If dimensions are not specified, full array reduction is performed (Size: AtLeast(min=0))

any

INDArray any(INDArray x, int[] dimensions)

SDVariable any(SDVariable x, int[] dimensions)
SDVariable any(String name, SDVariable x, int[] dimensions)

Boolean or array reduction operation, optionally along specified dimensions

  • x (NDARRAY) - Input variable

  • dimensions - Dimensions to reduce over. If dimensions are not specified, full array reduction is performed (Size: AtLeast(min=0))

argmax

INDArray argmax(INDArray in, boolean keepDims, int[] dimensions)
INDArray argmax(INDArray in, int[] dimensions)

SDVariable argmax(SDVariable in, boolean keepDims, int[] dimensions)
SDVariable argmax(SDVariable in, int[] dimensions)
SDVariable argmax(String name, SDVariable in, boolean keepDims, int[] dimensions)
SDVariable argmax(String name, SDVariable in, int[] dimensions)

Argmax array reduction operation, optionally along specified dimensions.

Output values are the index of the maximum value of each slice along the specified dimension.

Note that if keepDims = true, the output variable has the same rank as the input variable,

with the reduced dimensions having size 1. This can be useful for later broadcast operations (such as subtracting

the mean along a dimension).

Example: if input has shape [a,b,c] and dimensions=[1] then output has shape:

keepDims = true: [a,1,c]

keepDims = false: [a,c]

  • in (NUMERIC) - Input variable

  • keepDims - If true: keep the dimensions that are reduced on (as size 1). False: remove the reduction dimensions - default = false

  • dimensions - Dimensions to reduce over. If dimensions are not specified, full array reduction is performed (Size: AtLeast(min=0))

argmin

INDArray argmin(INDArray in, boolean keepDims, int[] dimensions)
INDArray argmin(INDArray in, int[] dimensions)

SDVariable argmin(SDVariable in, boolean keepDims, int[] dimensions)
SDVariable argmin(SDVariable in, int[] dimensions)
SDVariable argmin(String name, SDVariable in, boolean keepDims, int[] dimensions)
SDVariable argmin(String name, SDVariable in, int[] dimensions)

Argmin array reduction operation, optionally along specified dimensions.

Output values are the index of the minimum value of each slice along the specified dimension.

Note that if keepDims = true, the output variable has the same rank as the input variable,

with the reduced dimensions having size 1. This can be useful for later broadcast operations (such as subtracting

the mean along a dimension).

Example: if input has shape [a,b,c] and dimensions=[1] then output has shape:

keepDims = true: [a,1,c]

keepDims = false: [a,c]

Note: supports broadcasting if x and y have different shapes and are broadcastable.

For example, if X has shape [1,10] and Y has shape [5,10] then op(X,Y) has output shape [5,10]

Broadcast rules are the same as NumPy: https://docs.scipy.org/doc/numpy/user/basics.broadcasting.html

  • in (NUMERIC) - Input variable

  • keepDims - If true: keep the dimensions that are reduced on (as size 1). False: remove the reduction dimensions - default = false

  • dimensions - Dimensions to reduce over. If dimensions are not specified, full array reduction is performed (Size: AtLeast(min=0))

batchMmul

INDArray batchMmul(INDArray inputsA, INDArray inputsB, boolean transposeA, boolean transposeB)
INDArray batchMmul(INDArray inputsA, INDArray inputsB)

SDVariable batchMmul(SDVariable inputsA, SDVariable inputsB, boolean transposeA, boolean transposeB)
SDVariable batchMmul(SDVariable inputsA, SDVariable inputsB)
SDVariable batchMmul(String name, SDVariable inputsA, SDVariable inputsB, boolean transposeA, boolean transposeB)
SDVariable batchMmul(String name, SDVariable inputsA, SDVariable inputsB)

Matrix multiply a batch of matrices. matricesA and matricesB have to be arrays of same

length and each pair taken from these sets has to have dimensions (M, N) and (N, K),

respectively. If transposeA is true, matrices from matricesA will have shape (N, M) instead.

Likewise, if transposeB is true, matrices from matricesB will have shape (K, N).

The result of this operation will be a batch of multiplied matrices. The

result has the same length as both input batches and each output matrix is of shape (M, K).

  • inputsA (NUMERIC) - First array of input matrices, all of shape (M, N) or (N, M)

  • inputsB (NUMERIC) - Second array of input matrices, all of shape (N, K) or (K, N)

  • transposeA - Whether to transpose A arrays or not - default = false

  • transposeB - Whether to transpose B arrays or not - default = false

castTo

INDArray castTo(INDArray arg, DataType datatype)

SDVariable castTo(SDVariable arg, DataType datatype)
SDVariable castTo(String name, SDVariable arg, DataType datatype)

Cast the array to a new datatype - for example, Integer -> Float

  • arg (NDARRAY) - Input variable to cast

  • datatype - Datatype to cast to

concat

INDArray concat(INDArray inputs, int dimension)

SDVariable concat(SDVariable inputs, int dimension)
SDVariable concat(String name, SDVariable inputs, int dimension)

Concatenate a set of inputs along the specified dimension.

Note that inputs must have identical rank and identical dimensions, other than the dimension to stack on.

For example, if 2 inputs have shape [a, x, c] and [a, y, c] and dimension = 1, then the output has shape [a, x+y, c]

  • inputs (NUMERIC) - Input variables

  • dimension - Dimension to concatenate on

cumprod

INDArray cumprod(INDArray in, boolean exclusive, boolean reverse, int[] axis)
INDArray cumprod(INDArray in, int[] axis)

SDVariable cumprod(SDVariable in, boolean exclusive, boolean reverse, int[] axis)
SDVariable cumprod(SDVariable in, int[] axis)
SDVariable cumprod(String name, SDVariable in, boolean exclusive, boolean reverse, int[] axis)
SDVariable cumprod(String name, SDVariable in, int[] axis)

Cumulative product operation.

For input: [ a, b, c], output is:

exclusive=false, reverse=false: [a, a_b, a_b*c]

exclusive=true, reverse=false, [0, a, a*b]

exclusive=false, reverse=true: [a_b_c, b*c, c]

exclusive=true, reverse=true: [b*c, c, 0]

  • in (NUMERIC) - Input variable

  • exclusive - If true: exclude the first value - default = false

  • reverse - If true: reverse the direction of the accumulation - default = false

  • axis - Scalar axis argument for dimension to perform cumululative sum operations along (Size: AtLeast(min=1))

cumsum

INDArray cumsum(INDArray in, boolean exclusive, boolean reverse, int[] axis)
INDArray cumsum(INDArray in, int[] axis)

SDVariable cumsum(SDVariable in, boolean exclusive, boolean reverse, int[] axis)
SDVariable cumsum(SDVariable in, int[] axis)
SDVariable cumsum(String name, SDVariable in, boolean exclusive, boolean reverse, int[] axis)
SDVariable cumsum(String name, SDVariable in, int[] axis)

Cumulative sum operation.

For input: [ a, b, c], output is:

exclusive=false, reverse=false: [a, a+b, a+b+c]

exclusive=true, reverse=false, [0, a, a+b]

exclusive=false, reverse=true: [a+b+c, b+c, c]

exclusive=true, reverse=true: [b+c, c, 0]

  • in (NUMERIC) - Input variable

  • exclusive - If true: exclude the first value - default = false

  • reverse - If true: reverse the direction of the accumulation - default = false

  • axis - Scalar axis argument for dimension to perform cumululative sum operations along (Size: AtLeast(min=1))

dot

INDArray dot(INDArray x, INDArray y, int[] dimensions)

SDVariable dot(SDVariable x, SDVariable y, int[] dimensions)
SDVariable dot(String name, SDVariable x, SDVariable y, int[] dimensions)

Pairwise dot product reduction along dimension

output = sum(i=0 ... size(dim)-1) x[i] * y[i]

  • x (NUMERIC) - first input

  • y (NUMERIC) - second input

  • dimensions - Dimensions to reduce over. If dimensions are not specified, full array reduction is performed (Size: AtLeast(min=0))

dynamicPartition

INDArray dynamicPartition(INDArray x, INDArray partitions, int numPartitions)

SDVariable dynamicPartition(SDVariable x, SDVariable partitions, int numPartitions)
SDVariable dynamicPartition(String name, SDVariable x, SDVariable partitions, int numPartitions)

Dynamically partition the input variable values into the specified number of paritions, using the indices.

Example:

input = [1,2,3,4,5]

numPartitions = 2

partitions = [1,0,0,1,0]

out[0] = [2,3,5]

out[1] = [1,4] `
  • x (NUMERIC) - Input variable

  • partitions (INT) - 1D input with values 0 to numPartitions-1

  • numPartitions - Number of partitions, >= 1

dynamicStitch

INDArray dynamicStitch(INDArray indices, INDArray x)

SDVariable dynamicStitch(SDVariable indices, SDVariable x)
SDVariable dynamicStitch(String name, SDVariable indices, SDVariable x)

Dynamically merge the specified input arrays into a single array, using the specified indices

  • indices (INT) - Indices to use when merging. Must be >= 1, same length as input variables

  • x (NUMERIC) - Input variables.

eq

INDArray eq(INDArray x, double y)

SDVariable eq(SDVariable x, double y)
SDVariable eq(String name, SDVariable x, double y)

Equals operation: elementwise x == y

Return boolean array with values true where satisfied, or false otherwise.

  • x (NUMERIC) - Input array

  • y - Double value argument to use in operation

eq

INDArray eq(INDArray x, INDArray y)

SDVariable eq(SDVariable x, SDVariable y)
SDVariable eq(String name, SDVariable x, SDVariable y)

Equal to operation: elementwise x == y

If x and y arrays have equal shape, the output shape is the same as these inputs.

Note: supports broadcasting if x and y have different shapes and are broadcastable.

For example, if X has shape [1,10] and Y has shape [5,10] then op(X,Y) has output shape [5,10]

Broadcast rules are the same as NumPy: https://docs.scipy.org/doc/numpy/user/basics.broadcasting.html

Return boolean array with values true where satisfied, or false otherwise.

  • x (NUMERIC) - Input 1

  • y (NUMERIC) - Input 2

expandDims

INDArray expandDims(INDArray x, int axis)

SDVariable expandDims(SDVariable x, int axis)
SDVariable expandDims(String name, SDVariable x, int axis)

Reshape the input by adding a 1 at the specified location.

For example, if input has shape [a, b], then output shape is:

axis = 0: [1, a, b]

axis = 1: [a, 1, b]

axis = 2: [a, b, 1]

  • x (NDARRAY) - Input variable

  • axis - Axis to expand

fill

INDArray fill(INDArray shape, DataType dataType, double value)

SDVariable fill(SDVariable shape, DataType dataType, double value)
SDVariable fill(String name, SDVariable shape, DataType dataType, double value)

Generate an output variable with the specified (dynamic) shape with all elements set to the specified value

  • shape (INT) - Shape: must be a 1D array/variable

  • dataType - Datatype of the output array

  • value - Value to set all elements to

gather

INDArray gather(INDArray df, int[] indices, int axis)

SDVariable gather(SDVariable df, int[] indices, int axis)
SDVariable gather(String name, SDVariable df, int[] indices, int axis)

Gather slices from the input variable where the indices are specified as fixed int[] values.

Output shape is same as input shape, except for axis dimension, which has size equal to indices.length.

  • df (NUMERIC) - Input variable

  • indices - Indices to get (Size: AtLeast(min=1))

  • axis - Axis that the indices refer to

gather

INDArray gather(INDArray df, INDArray indices, int axis)

SDVariable gather(SDVariable df, SDVariable indices, int axis)
SDVariable gather(String name, SDVariable df, SDVariable indices, int axis)

Gather slices from the input variable where the indices are specified as dynamic array values.

Output shape is same as input shape, except for axis dimension, which has size equal to indices.length.

  • df (NUMERIC) - Input variable

  • indices (INT) - Indices to get slices for. Rank 0 or 1 input

  • axis - Axis that the indices refer to

gatherNd

INDArray gatherNd(INDArray df, INDArray indices)

SDVariable gatherNd(SDVariable df, SDVariable indices)
SDVariable gatherNd(String name, SDVariable df, SDVariable indices)

Gather slices from df with shape specified by indices.

  • df (NUMERIC) -

  • indices (NUMERIC) -

gt

INDArray gt(INDArray x, double y)

SDVariable gt(SDVariable x, double y)
SDVariable gt(String name, SDVariable x, double y)

Greater than operation: elementwise x > y

Return boolean array with values true where satisfied, or false otherwise.

  • x (NUMERIC) - Input array

  • y - Double value argument to use in operation

gt

INDArray gt(INDArray x, INDArray y)

SDVariable gt(SDVariable x, SDVariable y)
SDVariable gt(String name, SDVariable x, SDVariable y)

Greater than operation: elementwise x > y

If x and y arrays have equal shape, the output shape is the same as these inputs.

Note: supports broadcasting if x and y have different shapes and are broadcastable.

For example, if X has shape [1,10] and Y has shape [5,10] then op(X,Y) has output shape [5,10]

Broadcast rules are the same as NumPy: https://docs.scipy.org/doc/numpy/user/basics.broadcasting.html

Return boolean array with values true where satisfied, or false otherwise.

  • x (NUMERIC) - Input 1

  • y (NUMERIC) - Input 2

gte

INDArray gte(INDArray x, double y)

SDVariable gte(SDVariable x, double y)
SDVariable gte(String name, SDVariable x, double y)

Greater than or equals operation: elementwise x >= y

Return boolean array with values true where satisfied, or false otherwise.

  • x (NUMERIC) - Input array

  • y - Double value argument to use in operation

gte

INDArray gte(INDArray x, INDArray y)

SDVariable gte(SDVariable x, SDVariable y)
SDVariable gte(String name, SDVariable x, SDVariable y)

Greater than or equal to operation: elementwise x >= y

If x and y arrays have equal shape, the output shape is the same as these inputs.

Note: supports broadcasting if x and y have different shapes and are broadcastable.

For example, if X has shape [1,10] and Y has shape [5,10] then op(X,Y) has output shape [5,10]

Broadcast rules are the same as NumPy: https://docs.scipy.org/doc/numpy/user/basics.broadcasting.html

Return boolean array with values true where satisfied, or false otherwise.

  • x (NUMERIC) - Input 1

  • y (NUMERIC) - Input 2

identity

INDArray identity(INDArray input)

SDVariable identity(SDVariable input)
SDVariable identity(String name, SDVariable input)

Elementwise identity operation: out = x

  • input (NUMERIC) - Input variable

invertPermutation

INDArray invertPermutation(INDArray input)

SDVariable invertPermutation(SDVariable input)
SDVariable invertPermutation(String name, SDVariable input)

Compute the inverse permutation indices for a permutation operation

Example: if input is [2, 0, 1] then output is [1, 2, 0]

The idea is that x.permute(input).permute(invertPermutation(input)) == x

  • input (INT) - 1D indices for permutation

isNumericTensor

INDArray isNumericTensor(INDArray x)

SDVariable isNumericTensor(SDVariable x)
SDVariable isNumericTensor(String name, SDVariable x)

Is the director a numeric tensor? In the current version of ND4J/SameDiff, this always returns true/1

  • x (NUMERIC) - Input variable

linspace

INDArray linspace(DataType dataType, double start, double stop, long number)

SDVariable linspace(DataType dataType, double start, double stop, long number)
SDVariable linspace(String name, DataType dataType, double start, double stop, long number)

Create a new 1d array with values evenly spaced between values 'start' and 'stop'

For example, linspace(start=3.0, stop=4.0, number=3) will generate [3.0, 3.5, 4.0]

  • dataType - Data type of the output array

  • start - Start value

  • stop - Stop value

  • number - Number of values to generate

linspace

INDArray linspace(INDArray start, INDArray stop, INDArray number, DataType dataType)

SDVariable linspace(SDVariable start, SDVariable stop, SDVariable number, DataType dataType)
SDVariable linspace(String name, SDVariable start, SDVariable stop, SDVariable number, DataType dataType)

Create a new 1d array with values evenly spaced between values 'start' and 'stop'

For example, linspace(start=3.0, stop=4.0, number=3) will generate [3.0, 3.5, 4.0]

  • start (NUMERIC) - Start value

  • stop (NUMERIC) - Stop value

  • number (LONG) - Number of values to generate

  • dataType - Data type of the output array

lt

INDArray lt(INDArray x, double y)

SDVariable lt(SDVariable x, double y)
SDVariable lt(String name, SDVariable x, double y)

Less than operation: elementwise x < y

Return boolean array with values true where satisfied, or false otherwise.

  • x (NUMERIC) - Input array

  • y - Double value argument to use in operation

lt

INDArray lt(INDArray x, INDArray y)

SDVariable lt(SDVariable x, SDVariable y)
SDVariable lt(String name, SDVariable x, SDVariable y)

Less than operation: elementwise x < y

If x and y arrays have equal shape, the output shape is the same as these inputs.

Note: supports broadcasting if x and y have different shapes and are broadcastable.

For example, if X has shape [1,10] and Y has shape [5,10] then op(X,Y) has output shape [5,10]

Broadcast rules are the same as NumPy: https://docs.scipy.org/doc/numpy/user/basics.broadcasting.html

Return boolean array with values true where satisfied, or false otherwise.

  • x (NUMERIC) - Input 1

  • y (NUMERIC) - Input 2

lte

INDArray lte(INDArray x, double y)

SDVariable lte(SDVariable x, double y)
SDVariable lte(String name, SDVariable x, double y)

Less than or equals operation: elementwise x <= y

Return boolean array with values true where satisfied, or false otherwise.

  • x (NUMERIC) - Input array

  • y - Double value argument to use in operation

lte

INDArray lte(INDArray x, INDArray y)

SDVariable lte(SDVariable x, SDVariable y)
SDVariable lte(String name, SDVariable x, SDVariable y)

Less than or equal to operation: elementwise x <= y

If x and y arrays have equal shape, the output shape is the same as these inputs.

Note: supports broadcasting if x and y have different shapes and are broadcastable.

For example, if X has shape [1,10] and Y has shape [5,10] then op(X,Y) has output shape [5,10]

Broadcast rules are the same as NumPy: https://docs.scipy.org/doc/numpy/user/basics.broadcasting.html

Return boolean array with values true where satisfied, or false otherwise.

  • x (NUMERIC) - Input 1

  • y (NUMERIC) - Input 2

matchCondition

INDArray matchCondition(INDArray in, Condition condition)

SDVariable matchCondition(SDVariable in, Condition condition)
SDVariable matchCondition(String name, SDVariable in, Condition condition)

Returns a boolean mask of equal shape to the input, where the condition is satisfied - value 1 where satisfied, 0 otherwise

  • in (NUMERIC) - Input

  • condition - Condition

matchConditionCount

INDArray matchConditionCount(INDArray in, Condition condition)

SDVariable matchConditionCount(SDVariable in, Condition condition)
SDVariable matchConditionCount(String name, SDVariable in, Condition condition)

Returns a count of the number of elements that satisfy the condition

  • in (NUMERIC) - Input

  • condition - Condition

matchConditionCount

INDArray matchConditionCount(INDArray in, Condition condition, boolean keepDim, int[] dimensions)
INDArray matchConditionCount(INDArray in, Condition condition, int[] dimensions)

SDVariable matchConditionCount(SDVariable in, Condition condition, boolean keepDim, int[] dimensions)
SDVariable matchConditionCount(SDVariable in, Condition condition, int[] dimensions)
SDVariable matchConditionCount(String name, SDVariable in, Condition condition, boolean keepDim, int[] dimensions)
SDVariable matchConditionCount(String name, SDVariable in, Condition condition, int[] dimensions)

Returns a count of the number of elements that satisfy the condition (for each slice along the specified dimensions)

Note that if keepDims = true, the output variable has the same rank as the input variable,

with the reduced dimensions having size 1. This can be useful for later broadcast operations (such as subtracting

the mean along a dimension).

Example: if input has shape [a,b,c] and dimensions=[1] then output has shape:

keepDims = true: [a,1,c]

keepDims = false: [a,c]

  • in (NUMERIC) - Input variable

  • condition - Condition

  • keepDim - If true: keep the dimensions that are reduced on (as size 1). False: remove the reduction dimensions - default = false

  • dimensions - Dimensions to reduce over. If dimensions are not specified, full array reduction is performed (Size: AtLeast(min=0))

max

INDArray max(INDArray x, boolean keepDims, int[] dimensions)
INDArray max(INDArray x, int[] dimensions)

SDVariable max(SDVariable x, boolean keepDims, int[] dimensions)
SDVariable max(SDVariable x, int[] dimensions)
SDVariable max(String name, SDVariable x, boolean keepDims, int[] dimensions)
SDVariable max(String name, SDVariable x, int[] dimensions)

Max array reduction operation, optionally along specified dimensions

Note that if keepDims = true, the output variable has the same rank as the input variable,

with the reduced dimensions having size 1. This can be useful for later broadcast operations (such as subtracting

the mean along a dimension).

Example: if input has shape [a,b,c] and dimensions=[1] then output has shape:

keepDims = true: [a,1,c]

keepDims = false: [a,c]

  • x (NUMERIC) - Input variable

  • keepDims - If true: keep the dimensions that are reduced on (as size 1). False: remove the reduction dimensions - default = false

  • dimensions - Dimensions to reduce over. If dimensions are not specified, full array reduction is performed (Size: AtLeast(min=0))

max

INDArray max(INDArray first, INDArray second)

SDVariable max(SDVariable first, SDVariable second)
SDVariable max(String name, SDVariable first, SDVariable second)

Element-wise maximum operation: out[i] = max(first[i], second[i])

Note: supports broadcasting if x and y have different shapes and are broadcastable.

For example, if X has shape [1,10] and Y has shape [5,10] then op(X,Y) has output shape [5,10]

Broadcast rules are the same as NumPy: https://docs.scipy.org/doc/numpy/user/basics.broadcasting.html

  • first (NUMERIC) - First input array

  • second (NUMERIC) - Second input array

mean

INDArray mean(INDArray x, boolean keepDims, int[] dimensions)
INDArray mean(INDArray x, int[] dimensions)

SDVariable mean(SDVariable x, boolean keepDims, int[] dimensions)
SDVariable mean(SDVariable x, int[] dimensions)
SDVariable mean(String name, SDVariable x, boolean keepDims, int[] dimensions)
SDVariable mean(String name, SDVariable x, int[] dimensions)

Mean (average) array reduction operation, optionally along specified dimensions

Note that if keepDims = true, the output variable has the same rank as the input variable,

with the reduced dimensions having size 1. This can be useful for later broadcast operations (such as subtracting

the mean along a dimension).

Example: if input has shape [a,b,c] and dimensions=[1] then output has shape:

keepDims = true: [a,1,c]

keepDims = false: [a,c]

  • x (NUMERIC) - Input variable

  • keepDims - If true: keep the dimensions that are reduced on (as size 1). False: remove the reduction dimensions - default = false

  • dimensions - Dimensions to reduce over. If dimensions are not specified, full array reduction is performed (Size: AtLeast(min=0))

merge

INDArray merge(INDArray x, INDArray y)

SDVariable merge(SDVariable x, SDVariable y)
SDVariable merge(String name, SDVariable x, SDVariable y)

The merge operation is a control operation that forwards the either of the inputs to the output, when

the first of them becomes available. If both are available, the output is undefined (either input could

be forwarded to the output)

  • x (NUMERIC) - Input variable

  • y (NUMERIC) - Input variable

min

INDArray min(INDArray x, boolean keepDims, int[] dimensions)
INDArray min(INDArray x, int[] dimensions)

SDVariable min(SDVariable x, boolean keepDims, int[] dimensions)
SDVariable min(SDVariable x, int[] dimensions)
SDVariable min(String name, SDVariable x, boolean keepDims, int[] dimensions)
SDVariable min(String name, SDVariable x, int[] dimensions)

Minimum array reduction operation, optionally along specified dimensions. out = min(in)

Note that if keepDims = true, the output variable has the same rank as the input variable,

with the reduced dimensions having size 1. This can be useful for later broadcast operations (such as subtracting

the mean along a dimension).

Example: if input has shape [a,b,c] and dimensions=[1] then output has shape:

keepDims = true: [a,1,c]

keepDims = false: [a,c]

  • x (NUMERIC) - Input variable

  • keepDims - If true: keep the dimensions that are reduced on (as size 1). False: remove the reduction dimensions - default = false

  • dimensions - Dimensions to reduce over. If dimensions are not specified, full array reduction is performed (Size: AtLeast(min=0))

min

INDArray min(INDArray first, INDArray second)

SDVariable min(SDVariable first, SDVariable second)
SDVariable min(String name, SDVariable first, SDVariable second)

Element-wise minimum operation: out[i] = min(first[i], second[i])

Note: supports broadcasting if x and y have different shapes and are broadcastable.

For example, if X has shape [1,10] and Y has shape [5,10] then op(X,Y) has output shape [5,10]

Broadcast rules are the same as NumPy: https://docs.scipy.org/doc/numpy/user/basics.broadcasting.html

  • first (NUMERIC) - First input array

  • second (NUMERIC) - Second input array

mmul

INDArray mmul(INDArray x, INDArray y, boolean transposeX, boolean transposeY, boolean transposeZ)
INDArray mmul(INDArray x, INDArray y)

SDVariable mmul(SDVariable x, SDVariable y, boolean transposeX, boolean transposeY, boolean transposeZ)
SDVariable mmul(SDVariable x, SDVariable y)
SDVariable mmul(String name, SDVariable x, SDVariable y, boolean transposeX, boolean transposeY, boolean transposeZ)
SDVariable mmul(String name, SDVariable x, SDVariable y)

Matrix multiplication: out = mmul(x,y)

Supports specifying transpose argument to perform operation such as mmul(a^T, b), etc.

  • x (NUMERIC) - First input variable

  • y (NUMERIC) - Second input variable

  • transposeX - Transpose x (first argument) - default = false

  • transposeY - Transpose y (second argument) - default = false

  • transposeZ - Transpose result array - default = false

neq

INDArray neq(INDArray x, double y)

SDVariable neq(SDVariable x, double y)
SDVariable neq(String name, SDVariable x, double y)

Not equals operation: elementwise x != y

Return boolean array with values true where satisfied, or false otherwise.

  • x (NUMERIC) - Input array

  • y - Double value argument to use in operation

neq

INDArray neq(INDArray x, INDArray y)

SDVariable neq(SDVariable x, SDVariable y)
SDVariable neq(String name, SDVariable x, SDVariable y)

Not equal to operation: elementwise x != y

If x and y arrays have equal shape, the output shape is the same as these inputs.

Note: supports broadcasting if x and y have different shapes and are broadcastable.

For example, if X has shape [1,10] and Y has shape [5,10] then op(X,Y) has output shape [5,10]

Broadcast rules are the same as NumPy: https://docs.scipy.org/doc/numpy/user/basics.broadcasting.html

Return boolean array with values true where satisfied, or false otherwise.

  • x (NUMERIC) - Input 1

  • y (NUMERIC) - Input 2

norm1

INDArray norm1(INDArray x, boolean keepDims, int[] dimensions)
INDArray norm1(INDArray x, int[] dimensions)

SDVariable norm1(SDVariable x, boolean keepDims, int[] dimensions)
SDVariable norm1(SDVariable x, int[] dimensions)
SDVariable norm1(String name, SDVariable x, boolean keepDims, int[] dimensions)
SDVariable norm1(String name, SDVariable x, int[] dimensions)

Norm1 (L1 norm) reduction operation: The output contains the L1 norm for each tensor/subset along the specified dimensions:

out = sum_i abs(x[i])

Note that if keepDims = true, the output variable has the same rank as the input variable,

with the reduced dimensions having size 1. This can be useful for later broadcast operations (such as subtracting

the mean along a dimension).

Example: if input has shape [a,b,c] and dimensions=[1] then output has shape:

keepDims = true: [a,1,c]

keepDims = false: [a,c]

  • x (NUMERIC) - Input variable

  • keepDims - If true: keep the dimensions that are reduced on (as size 1). False: remove the reduction dimensions - default = false

  • dimensions - dimensions to reduce over (Size: AtLeast(min=0))

norm2

INDArray norm2(INDArray x, boolean keepDims, int[] dimensions)
INDArray norm2(INDArray x, int[] dimensions)

SDVariable norm2(SDVariable x, boolean keepDims, int[] dimensions)
SDVariable norm2(SDVariable x, int[] dimensions)
SDVariable norm2(String name, SDVariable x, boolean keepDims, int[] dimensions)
SDVariable norm2(String name, SDVariable x, int[] dimensions)

Norm2 (L2 norm) reduction operation: The output contains the L2 norm for each tensor/subset along the specified dimensions:

out = sqrt(sum_i x[i]^2)

Note that if keepDims = true, the output variable has the same rank as the input variable,

with the reduced dimensions having size 1. This can be useful for later broadcast operations (such as subtracting

the mean along a dimension).

Example: if input has shape [a,b,c] and dimensions=[1] then output has shape:

keepDims = true: [a,1,c]

keepDims = false: [a,c]

  • x (NUMERIC) - Input variable

  • keepDims - If true: keep the dimensions that are reduced on (as size 1). False: remove the reduction dimensions - default = false

  • dimensions - dimensions dimensions to reduce over (Size: AtLeast(min=0))

normmax

INDArray normmax(INDArray x, boolean keepDims, int[] dimensions)
INDArray normmax(INDArray x, int[] dimensions)

SDVariable normmax(SDVariable x, boolean keepDims, int[] dimensions)
SDVariable normmax(SDVariable x, int[] dimensions)
SDVariable normmax(String name, SDVariable x, boolean keepDims, int[] dimensions)
SDVariable normmax(String name, SDVariable x, int[] dimensions)

Max norm (infinity norm) reduction operation: The output contains the max norm for each tensor/subset along the

specified dimensions:

out = max(abs(x[i]))

Note that if keepDims = true, the output variable has the same rank as the input variable,

with the reduced dimensions having size 1. This can be useful for later broadcast operations (such as subtracting

the mean along a dimension).

Example: if input has shape [a,b,c] and dimensions=[1] then output has shape:

keepDims = true: [a,1,c]

keepDims = false: [a,c]

  • x (NUMERIC) - Input variable

  • keepDims - If true: keep the dimensions that are reduced on (as size 1). False: remove the reduction dimensions - default = false

  • dimensions - dimensions to reduce over (Size: AtLeast(min=0))

oneHot

INDArray oneHot(INDArray indices, int depth, int axis, double on, double off, DataType dataType)
INDArray oneHot(INDArray indices, int depth, int axis, double on, double off)

SDVariable oneHot(SDVariable indices, int depth, int axis, double on, double off, DataType dataType)
SDVariable oneHot(SDVariable indices, int depth, int axis, double on, double off)
SDVariable oneHot(String name, SDVariable indices, int depth, int axis, double on, double off, DataType dataType)
SDVariable oneHot(String name, SDVariable indices, int depth, int axis, double on, double off)

Convert the array to a one-hot array with walues and for each entry

If input has shape [ a, ..., n] then output has shape [ a, ..., n, depth],

with {out[i, ..., j, in[i,...,j]] with other values being set to

  • indices (NUMERIC) - Indices - value 0 to depth-1

  • depth - Number of classes

  • axis -

  • on -

  • off -

  • dataType - Output data type - default = DataType.FLOAT

oneHot

INDArray oneHot(INDArray indices, int depth)

SDVariable oneHot(SDVariable indices, int depth)
SDVariable oneHot(String name, SDVariable indices, int depth)

Convert the array to a one-hot array with walues 0 and 1 for each entry

If input has shape [ a, ..., n] then output has shape [ a, ..., n, depth],

with out[i, ..., j, in[i,...,j]] = 1 with other values being set to 0

see oneHot(SDVariable, int, int, double, double)

  • indices (NUMERIC) - Indices - value 0 to depth-1

  • depth - Number of classes

onesLike

INDArray onesLike(INDArray input)

SDVariable onesLike(SDVariable input)
SDVariable onesLike(String name, SDVariable input)

Return a variable of all 1s, with the same shape as the input variable. Note that this is dynamic:

if the input shape changes in later execution, the returned variable's shape will also be updated

  • input (NUMERIC) - Input INDArray

onesLike

INDArray onesLike(INDArray input, DataType dataType)

SDVariable onesLike(SDVariable input, DataType dataType)
SDVariable onesLike(String name, SDVariable input, DataType dataType)

As per onesLike(String, SDVariable) but the output datatype may be specified

  • input (NUMERIC) -

  • dataType -

permute

INDArray permute(INDArray x, INDArray dimensions)

SDVariable permute(SDVariable x, SDVariable dimensions)
SDVariable permute(String name, SDVariable x, SDVariable dimensions)

Array permutation operation: permute the dimensions according to the specified permutation indices.

Example: if input has shape [a,b,c] and dimensions = [2,0,1] the output has shape [c,a,b]

  • x (NUMERIC) - Input variable

  • dimensions (INT) - Permute dimensions

permute

INDArray permute(INDArray x, int[] dimensions)

SDVariable permute(SDVariable x, int[] dimensions)
SDVariable permute(String name, SDVariable x, int[] dimensions)

Array permutation operation: permute the dimensions according to the specified permutation indices.

Example: if input has shape [a,b,c] and dimensions = [2,0,1] the output has shape [c,a,b]

  • x (NUMERIC) - Input variable

  • dimensions - (Size: AtLeast(min=0))

prod

INDArray prod(INDArray x, boolean keepDims, int[] dimensions)
INDArray prod(INDArray x, int[] dimensions)

SDVariable prod(SDVariable x, boolean keepDims, int[] dimensions)
SDVariable prod(SDVariable x, int[] dimensions)
SDVariable prod(String name, SDVariable x, boolean keepDims, int[] dimensions)
SDVariable prod(String name, SDVariable x, int[] dimensions)

Product array reduction operation, optionally along specified dimensions

Note that if keepDims = true, the output variable has the same rank as the input variable,

with the reduced dimensions having size 1. This can be useful for later broadcast operations (such as subtracting

the mean along a dimension).

Example: if input has shape [a,b,c] and dimensions=[1] then output has shape:

keepDims = true: [a,1,c]

keepDims = false: [a,c]

  • x (NUMERIC) - Input variable

  • keepDims - If true: keep the dimensions that are reduced on (as size 1). False: remove the reduction dimensions - default = false

  • dimensions - Dimensions to reduce over. If dimensions are not specified, full array reduction is performed (Size: AtLeast(min=0))

range

INDArray range(double from, double to, double step, DataType dataType)

SDVariable range(double from, double to, double step, DataType dataType)
SDVariable range(String name, double from, double to, double step, DataType dataType)

Create a new variable with a 1d array, where the values start at from and increment by step

up to (but not including) limit.

For example, range(1.0, 3.0, 0.5) will return [1.0, 1.5, 2.0, 2.5]

  • from - Initial/smallest value

  • to - Largest value (exclusive)

  • step - Step size

  • dataType -

range

INDArray range(INDArray from, INDArray to, INDArray step, DataType dataType)

SDVariable range(SDVariable from, SDVariable to, SDVariable step, DataType dataType)
SDVariable range(String name, SDVariable from, SDVariable to, SDVariable step, DataType dataType)

Create a new variable with a 1d array, where the values start at from and increment by step

up to (but not including) limit.

For example, range(1.0, 3.0, 0.5) will return [1.0, 1.5, 2.0, 2.5]

  • from (NUMERIC) - Initial/smallest value

  • to (NUMERIC) - Largest value (exclusive)

  • step (NUMERIC) - Step size

  • dataType -

rank

INDArray rank(INDArray in)

SDVariable rank(SDVariable in)
SDVariable rank(String name, SDVariable in)

Returns the rank (number of dimensions, i.e., length(shape)) of the specified INDArray as a 0D scalar variable

  • in (NUMERIC) - Input variable

replaceWhere

INDArray replaceWhere(INDArray update, INDArray from, Condition condition)

SDVariable replaceWhere(SDVariable update, SDVariable from, Condition condition)
SDVariable replaceWhere(String name, SDVariable update, SDVariable from, Condition condition)

Element-wise replace where condition:

out[i] = from[i] if condition(update[i]) is satisfied, or

out[i] = update[i] if condition(update[i]) is NOT satisfied

  • update (NUMERIC) - Source array

  • from (NUMERIC) - Replacement values array (used conditionally). Must be same shape as 'update' array

  • condition - Condition to check on update array elements

replaceWhere

INDArray replaceWhere(INDArray update, double value, Condition condition)

SDVariable replaceWhere(SDVariable update, double value, Condition condition)
SDVariable replaceWhere(String name, SDVariable update, double value, Condition condition)

Element-wise replace where condition:

out[i] = value if condition(update[i]) is satisfied, or

out[i] = update[i] if condition(update[i]) is NOT satisfied

  • update (NUMERIC) - Source array

  • value - Value to set at the output, if the condition is satisfied

  • condition - Condition to check on update array elements

reshape

INDArray reshape(INDArray x, INDArray shape)

SDVariable reshape(SDVariable x, SDVariable shape)
SDVariable reshape(String name, SDVariable x, SDVariable shape)

Reshape the input variable to the specified (fixed) shape. The output variable will have the same values as the

input, but with the specified shape.

Note that prod(shape) must match length(input) == prod(input.shape)

  • x (NUMERIC) - Input variable

  • shape (NUMERIC) - New shape for variable

reshape

INDArray reshape(INDArray x, long[] shape)

SDVariable reshape(SDVariable x, long[] shape)
SDVariable reshape(String name, SDVariable x, long[] shape)

Reshape the input variable to the specified (fixed) shape. The output variable will have the same values as the

input, but with the specified shape.

Note that prod(shape) must match length(input) == prod(input.shape)

  • x (NUMERIC) - Input variable

  • shape - New shape for variable (Size: AtLeast(min=0))

reverse

INDArray reverse(INDArray x, int[] dimensions)

SDVariable reverse(SDVariable x, int[] dimensions)
SDVariable reverse(String name, SDVariable x, int[] dimensions)

Reverse the values of an array for the specified dimensions

If input is:

[ 1, 2, 3]

[ 4, 5, 6]

then

reverse(in, 0):

[3, 2, 1]

[6, 5, 4]

reverse(in, 1):

[4, 5, 6]

[1, 2 3]

  • x (NUMERIC) - Input variable

  • dimensions - Input variable (Size: AtLeast(min=0))

reverseSequence

INDArray reverseSequence(INDArray x, INDArray seq_lengths, int seqDim, int batchDim)
INDArray reverseSequence(INDArray x, INDArray seq_lengths)

SDVariable reverseSequence(SDVariable x, SDVariable seq_lengths, int seqDim, int batchDim)
SDVariable reverseSequence(SDVariable x, SDVariable seq_lengths)
SDVariable reverseSequence(String name, SDVariable x, SDVariable seq_lengths, int seqDim, int batchDim)
SDVariable reverseSequence(String name, SDVariable x, SDVariable seq_lengths)

Reverse sequence op: for each slice along dimension seqDimension, the first seqLength values are reversed

  • x (NUMERIC) - Input variable

  • seq_lengths (INT) - Length of the sequences

  • seqDim - Sequence dimension - default = -1

  • batchDim - Batch dimension - default = 0

scalarFloorMod

INDArray scalarFloorMod(INDArray in, double value)

SDVariable scalarFloorMod(SDVariable in, double value)
SDVariable scalarFloorMod(String name, SDVariable in, double value)

Element-wise scalar floor modulus operation: out = floorMod(in, value).

i.e., returns the remainder after division by 'value'

  • in (NUMERIC) - Input variable

  • value - Scalar value to compare

scalarMax

INDArray scalarMax(INDArray in, double value)

SDVariable scalarMax(SDVariable in, double value)
SDVariable scalarMax(String name, SDVariable in, double value)

Element-wise scalar maximum operation: out = max(in, value)

  • in (NUMERIC) - Input variable

  • value - Scalar value to compare

scalarMin

INDArray scalarMin(INDArray in, double value)

SDVariable scalarMin(SDVariable in, double value)
SDVariable scalarMin(String name, SDVariable in, double value)

Element-wise scalar minimum operation: out = min(in, value)

  • in (NUMERIC) - Input variable

  • value - Scalar value to compare

scalarSet

INDArray scalarSet(INDArray in, double set)

SDVariable scalarSet(SDVariable in, double set)
SDVariable scalarSet(String name, SDVariable in, double set)

Return a variable with equal shape to the input, but all elements set to value 'set'

  • in (NUMERIC) - Input variable

  • set - Value to set

scatterAdd

INDArray scatterAdd(INDArray ref, INDArray indices, INDArray updates)

SDVariable scatterAdd(SDVariable ref, SDVariable indices, SDVariable updates)
SDVariable scatterAdd(String name, SDVariable ref, SDVariable indices, SDVariable updates)

Scatter addition operation.

If indices is rank 0 (a scalar), then out[index, ...] = out[index, ...] + op(updates[...])

If indices is rank 1 (a vector), then for each position i, out[indices[i], ...] = out[indices[i], ...] + op(updates[i, ...])

If indices is rank 2+, then for each position (i,...,k), out[indices[i], ..., indices[k], ...] = out[indices[i], ..., indices[k], ...] + op(updates[i, ..., k, ...])

Note that if multiple indices refer to the same location, the contributions from each is handled correctly.

  • ref (NUMERIC) - Initial/source variable

  • indices (NUMERIC) - Indices array

  • updates (NUMERIC) - Updates to add to the initial/source array

scatterDiv

INDArray scatterDiv(INDArray ref, INDArray indices, INDArray updates)

SDVariable scatterDiv(SDVariable ref, SDVariable indices, SDVariable updates)
SDVariable scatterDiv(String name, SDVariable ref, SDVariable indices, SDVariable updates)

Scatter division operation.

If indices is rank 0 (a scalar), then out[index, ...] = out[index, ...] + op(updates[...])

If indices is rank 1 (a vector), then for each position i, out[indices[i], ...] = out[indices[i], ...] + op(updates[i, ...])

If indices is rank 2+, then for each position (i,...,k), out[indices[i], ..., indices[k], ...] = out[indices[i], ..., indices[k], ...] + op(updates[i, ..., k, ...])

Note that if multiple indices refer to the same location, the contributions from each is handled correctly.

  • ref (NUMERIC) - Initial/source variable

  • indices (NUMERIC) - Indices array

  • updates (NUMERIC) - Updates to add to the initial/source array

scatterMax

INDArray scatterMax(INDArray ref, INDArray indices, INDArray updates)

SDVariable scatterMax(SDVariable ref, SDVariable indices, SDVariable updates)
SDVariable scatterMax(String name, SDVariable ref, SDVariable indices, SDVariable updates)

Scatter max operation.

If indices is rank 0 (a scalar), then out[index, ...] = out[index, ...] + op(updates[...])

If indices is rank 1 (a vector), then for each position i, out[indices[i], ...] = out[indices[i], ...] + op(updates[i, ...])

If indices is rank 2+, then for each position (i,...,k), out[indices[i], ..., indices[k], ...] = out[indices[i], ..., indices[k], ...] + op(updates[i, ..., k, ...])

Note that if multiple indices refer to the same location, the contributions from each is handled correctly.

  • ref (NUMERIC) - Initial/source variable

  • indices (NUMERIC) - Indices array

  • updates (NUMERIC) - Updates to add to the initial/source array

scatterMin

INDArray scatterMin(INDArray ref, INDArray indices, INDArray updates)

SDVariable scatterMin(SDVariable ref, SDVariable indices, SDVariable updates)
SDVariable scatterMin(String name, SDVariable ref, SDVariable indices, SDVariable updates)

Scatter min operation.

If indices is rank 0 (a scalar), then out[index, ...] = out[index, ...] + op(updates[...])

If indices is rank 1 (a vector), then for each position i, out[indices[i], ...] = out[indices[i], ...] + op(updates[i, ...])

If indices is rank 2+, then for each position (i,...,k), out[indices[i], ..., indices[k], ...] = out[indices[i], ..., indices[k], ...] + op(updates[i, ..., k, ...])

Note that if multiple indices refer to the same location, the contributions from each is handled correctly.

  • ref (NUMERIC) - Initial/source variable

  • indices (NUMERIC) - Indices array

  • updates (NUMERIC) - Updates to add to the initial/source array

scatterMul

INDArray scatterMul(INDArray ref, INDArray indices, INDArray updates)

SDVariable scatterMul(SDVariable ref, SDVariable indices, SDVariable updates)
SDVariable scatterMul(String name, SDVariable ref, SDVariable indices, SDVariable updates)

Scatter multiplication operation.

If indices is rank 0 (a scalar), then out[index, ...] = out[index, ...] + op(updates[...])

If indices is rank 1 (a vector), then for each position i, out[indices[i], ...] = out[indices[i], ...] + op(updates[i, ...])

If indices is rank 2+, then for each position (i,...,k), out[indices[i], ..., indices[k], ...] = out[indices[i], ..., indices[k], ...] + op(updates[i, ..., k, ...])

Note that if multiple indices refer to the same location, the contributions from each is handled correctly.

  • ref (NUMERIC) - Initial/source variable

  • indices (NUMERIC) - Indices array

  • updates (NUMERIC) - Updates to add to the initial/source array

scatterSub

INDArray scatterSub(INDArray ref, INDArray indices, INDArray updates)

SDVariable scatterSub(SDVariable ref, SDVariable indices, SDVariable updates)
SDVariable scatterSub(String name, SDVariable ref, SDVariable indices, SDVariable updates)

Scatter subtraction operation.

If indices is rank 0 (a scalar), then out[index, ...] = out[index, ...] + op(updates[...])

If indices is rank 1 (a vector), then for each position i, out[indices[i], ...] = out[indices[i], ...] + op(updates[i, ...])

If indices is rank 2+, then for each position (i,...,k), out[indices[i], ..., indices[k], ...] = out[indices[i], ..., indices[k], ...] + op(updates[i, ..., k, ...])

Note that if multiple indices refer to the same location, the contributions from each is handled correctly.

  • ref (NUMERIC) - Initial/source variable

  • indices (NUMERIC) - Indices array

  • updates (NUMERIC) - Updates to add to the initial/source array

scatterUpdate

INDArray scatterUpdate(INDArray ref, INDArray indices, INDArray updates)

SDVariable scatterUpdate(SDVariable ref, SDVariable indices, SDVariable updates)
SDVariable scatterUpdate(String name, SDVariable ref, SDVariable indices, SDVariable updates)

Scatter update operation.

If indices is rank 0 (a scalar), then out[index, ...] = out[index, ...] + op(updates[...])

If indices is rank 1 (a vector), then for each position i, out[indices[i], ...] = out[indices[i], ...] + op(updates[i, ...])

If indices is rank 2+, then for each position (i,...,k), out[indices[i], ..., indices[k], ...] = out[indices[i], ..., indices[k], ...] + op(updates[i, ..., k, ...])

Note that if multiple indices refer to the same location, the contributions from each is handled correctly.

  • ref (NUMERIC) - Initial/source variable

  • indices (NUMERIC) - Indices array

  • updates (NUMERIC) - Updates to add to the initial/source array

segmentMax

INDArray segmentMax(INDArray data, INDArray segmentIds)