Losses
String Named Losses
OrderedDict({
'mse': nn.MSELoss,
'mae': nn.L1Loss,
'ce_loss': nn.CrossEntropyLoss,
'bce_loss': nn.BCEWithLogitsLoss,
})
Others
losses.CELoss()
It is an alternative to nn.CrossEntropy
, which integrates label smoothing.
loss_fn = keras4torch.losses.CELoss(label_smoothing=0.1)