BCEWithLogitsLoss

Also written: nn.BCEWithLogitsLoss, binary cross-entropy

The loss for a yes/no question. It takes one raw score (logit) per photo and float labels of 0 or 1, and its `pos_weight` option makes the rare class count more.