Contents

softMaxCrossEntropy(_:labels:axis:reuctionType:name:)

Creates a softmax cross-entropy loss operation and returns the result tensor.

Declaration

func softMaxCrossEntropy(_ sourceTensor: MPSGraphTensor, labels labelsTensor: MPSGraphTensor, axis: Int, reuctionType reductionType: MPSGraphLossReductionType, name: String?) -> MPSGraphTensor

Parameters

  • sourceTensor:

    The source tensor.

  • labelsTensor:

    The labels tensor.

  • axis:

    The axis over which the operation computes the softmax reduction.

  • reductionType:

    The type of reduction MPSGraph uses to reduce across all other axes than axis. See: Mpsgraphlossreductiontype.

  • name:

    The name for the operation.

Return Value

A valid MPSGraphTensor object.

Discussion

The softmax cross-entropy operation computes:

    loss = reduction( - labels*ln( softmax(source) )), where
    sotfmax(source) = exp(source) / sum( exp(source) ), and

the operation performs the reduction over the axis dimension.