Operations
Functions
avg Pool Grad
Link copied to clipboard
abstract fun avgPoolGrad(x: DTensor, poolHeight: Int, poolWidth: Int): DTensor
Content copied to clipboard
batch Norm
Link copied to clipboard
open fun batchNorm(input: DTensor, scaleShift: DTensor, derivativeId: DerivativeID): BatchNormResult
Content copied to clipboard
broadcast To
Link copied to clipboard
conv Impl
Link copied to clipboard
abstract fun convImpl(signal: DTensor, filter: DTensor, hStride: Int, vStride: Int, padding: Convolve.Padding2D, derivativeId: DerivativeID): DTensor
Content copied to clipboard
Applies convolution to the tensor signal, using filter. Both signal and filter must be of rank 4. The expected shape of signal is NHWC (num signal, height, width, channels) and the expected filter shape is OHWI (output channels, height, width, input channels) where C == I
gather At Indices
Link copied to clipboard
abstract fun gatherAtIndices(x: DTensor, indices: List<IntArray>): DTensor
Content copied to clipboard
identity Gradient Of Same Kind
Link copied to clipboard
abstract fun identityGradientOfSameKind(x: DTensor, halfShape: Shape): DTensor
Content copied to clipboard
if Then Else
Link copied to clipboard
abstract fun ifThenElse(condition: DTensor, whenTrue: DTensor, whenFalse: DTensor, derivativeId: DerivativeID): DTensor
Content copied to clipboard
log Softmax
Link copied to clipboard
log Softmax Grad
Link copied to clipboard
max Pool With Indices
Link copied to clipboard
outer Product
Link copied to clipboard
abstract fun outerProduct(x: DTensor, y: DTensor, derivativeId: DerivativeID): DTensor
Content copied to clipboard
reshape To Scalar
Link copied to clipboard
scatter At Indices
Link copied to clipboard
times Scalar
Link copied to clipboard
abstract fun timesScalar(left: DScalar, right: DTensor, derivativeId: DerivativeID): DTensor
Content copied to clipboard
unary Minus
Link copied to clipboard
zero Of Same Kind
Link copied to clipboard
Properties
Inheritors
TracingTensorOperations
Link copied to clipboard