oa::FnLoss::crossEntropyBwd

crossEntropyBwd: gradient w.r.t. logits: (softmax(logits) - onehot(targets)) / batch

FnC++ public headers

Function Documentation

Matrix oa::FnLoss::crossEntropyBwd(
	const Matrix & inLogits,
	const Matrix & inTargets
)

crossEntropyBwd: gradient w.r.t. logits: (softmax(logits) - onehot(targets)) / batch

inLogits

const Matrix &

[batch, classes] unnormalized logits

inTargets

const Matrix &

[batch] class indices (UInt8, UInt32, or non-negative Int32)

Matrix

The declared return value.