Why we require sigmoid activation over the BCE loss??

[kRiShNa-429407]

Video no : 77
Timestamp : 10:57

Why we need to use BCEWithLogits over simple BCE??

[LuluW8071]

BCEwithLogitsLoss is used because it has an in-built sigmoid activation function, allowing you to pass the raw model outputs directly while BCELoss expects the output of your model is within the range [0, 1] to represent probabilities.

You can also refer to pytorch documentation if you want more detailed explanation.