Hi everyone,
Beside CE loss, I want to use another loss after 30 epoch, so I use the “loss = 0*nn.some_loss(x,y)” before 30 epoch, why I get the same training data(loss, accuracy), but test acc decreases a lot?
Hi everyone,
Beside CE loss, I want to use another loss after 30 epoch, so I use the “loss = 0*nn.some_loss(x,y)” before 30 epoch, why I get the same training data(loss, accuracy), but test acc decreases a lot?