Class  Description 

Assign 
Assign op: x = y, with broadcast as required

ATan2 
Arc Tangent elementwise function

BatchToSpace 
Ndimensional batch to space operation.

BatchToSpaceND 
Ndimensional batch to space operation.

BitsHammingDistance  
BitwiseAnd 
Bitwise AND operation, broadcastable

BitwiseOr 
Bitwise OR operation, broadcastable

BitwiseXor 
Bitwise XOR operation, broadcastable

Choose 
This op allows us to (based on the passed in condition)
to return the element fulfilling the condition.

CReLU  
CReluBp  
CumProd  
CumSum 
Cumulative sum operation, optionally along dimension.

CyclicRShiftBits 
Elementwise roll operation, rolls bits to the left, <<

CyclicShiftBits 
Elementwise roll operation, rolls bits to the left, <<

Dilation2D 
Dilation2D op wrapper

DotProductAttention 
(optionally scaled) dot product attention
See also "Attention is all you need" (https://arxiv.org/abs/1706.03762, p. 4, eq. 1)

DotProductAttentionBp 
(optionally scaled) dot product attention Backprop
See also "Attention is all you need" (https://arxiv.org/abs/1706.03762, p. 4, eq. 1)

DynamicPartition 
Transforms a given input tensor into numPartitions partitions, as indicated by the indices in "partitions".

DynamicStitch 
Transforms a given input tensor into numPartitions partitions, as indicated by the indices in "partitions".

EqualTo 
Bit mask over the ndarrays as to whether
the components are equal or not

FakeQuantWithMinMaxArgs 
Fake quantization operation.

FakeQuantWithMinMaxVars 
Fake quantization operation.

Fill 
Fill an array of given "shape" with the provided "value", e.g.

GreaterThan 
Bit mask over the ndarrays as to whether
the components are greater than or not

GreaterThanOrEqual 
Bit mask over the ndarrays as to whether
the components are greater than or equal or not

InTopK 
In Top K op

InvertPermutation 
Inverse of index permutation.

IsNonDecreasing 
This op takes 1 ndimensional array as input,
and returns true if for every adjacent pair we have x[i] <= x[i+1].

IsNumericTensor 
This op takes 1 ndimensional array as input, and returns true if input is a numeric array.

IsStrictlyIncreasing 
This op takes 1 ndimensional array as input,
and returns true if for every adjacent pair we have x[i] < x[i+1].

LayerNorm 
Composed op: g*standarize(x) + b
Bias is optional, and can be set as null

LayerNormBp 
Composed op: g*standarize(x) + b
Bias is optional, and can be set as null

LessThan 
Bit mask over the ndarrays as to whether
the components are less than or not

LessThanOrEqual 
Bit mask over the ndarrays as to whether
the components are less than or equal or not

ListDiff  
LogicalAnd  
LogicalNot  
LogicalOr  
LogicalXor  
LogMatrixDeterminant 
Log Matrix Determinant op
Given input with shape [..., N, N] output the log determinant for each submatrix.

LogSoftMax 
Log(softmax(X))

MatrixDeterminant 
Matrix Determinant op
Given input with shape [..., N, N] output the determinant for each submatrix.

MatrixDiag  
MatrixDiagPart  
MatrixInverse 
Matrix Inverse Function

MatrixSetDiag  
Max 
Max function

MaximumBp  
Min 
Min function

MirrorPad  
MultiHeadDotProductAttention 
(optionally scaled) multi head dot product attention
See also "Attention is all you need" (https://arxiv.org/abs/1706.03762, pp. 4,5, "3.2.2 MultiHead Attention")

MultiHeadDotProductAttentionBp 
(optionally scaled) multi head dot product attention Backprop
See also "Attention is all you need" (https://arxiv.org/abs/1706.03762, pp. 4,5, "3.2.2 MultiHead Attention")

NotEqualTo 
Not equal to function:
Bit mask over whether 2 elements are not equal or not

ParallelConcat  
Pow 
Broadcastable elementwise power operation: x[i]^y[i]

Qr  
Reverse  
ReverseBp  
ReverseSequence 
Created by farizrahman4u on 3/16/18.

ReverseV2 
This is compatibility op for ReverseV2

RShiftBits 
Elementwise shift operation, shift bits to the right, >>

ShiftBits 
Elementwise shift operation, shift bits to the left, <<

SoftMax 
Soft max function
row_maxes is a row vector (max for each row)
row_maxes = rowmaxes(input)
diff = exp(input  max) / diff.rowSums()
Outputs a probability distribution.

SpaceToBatch 
Ndimensional space to batch operation.

SpaceToBatchND 
Ndimensional space to batch operation.

Standardize  
StandardizeBp  
Svd 
SVD  singular value decomposition

ThresholdRelu 
Threshold ReLU op.

TopK 
Top K op

Trace 
Matrix trace operation

Unique  
UniqueWithCounts  
XwPlusB 
Composed op: mmul (X, W) + b

Zeta 
Elementwise Zeta function.

Copyright © 2020. All rights reserved.