medium10 pts · first answer countspytorchloss-functions

Softmax before CrossEntropyLoss

What is wrong with this classification loss code?

probs = torch.softmax(logits, dim=1)
loss = nn.CrossEntropyLoss()(probs, targets)  # targets: class indices

A.CrossEntropyLoss applies log-softmax itself, so pass raw logitsB.Nothing, applying softmax first is the recommended patternC.The targets must be one-hot float vectors for this lossD.The softmax must use dim=0 when working with batches

Discussion 0

Join free to post in the discussion.

Softmax before CrossEntropyLoss

Discussion 0

Related questions