About the sigmoid activation funcion about pips HOT 2 CLOSED

Hi, although I'm not the author, I may be able to answer your question. The sigmoid exists as F.binary_cross_entropy_with_logits = F.sigmoid + F.binary_cross_entropy, which seems a trick to avoid numerical unstable. A more popular and exactly same thing is CrossEntropyLoss, which combines LogSoftmax() and NLLLoss() so that you won't see softmax, as pointed out here https://stackoverflow.com/questions/55675345/should-i-use-softmax-as-output-when-using-cross-entropy-loss-in-pytorch

from pips.

Hi, although I'm not the author, I may be able to answer your question. The sigmoid exists as F.binary_cross_entropy_with_logits = F.sigmoid + F.binary_cross_entropy, which seems a trick to avoid numerical unstable. A more popular and exactly same thing is CrossEntropyLoss, which combines LogSoftmax() and NLLLoss() so that you won't see softmax, as pointed out here https://stackoverflow.com/questions/55675345/should-i-use-softmax-as-output-when-using-cross-entropy-loss-in-pytorch

Thank you very much for your detailed reply! It really helps me a lot, thx!

from pips.