Questions about train.log
Closed this issue · 1 comments
Hello,
I was wondering in the train.log what loss, Closs, Eloss, wd, waet and top1avg mean?
Thanks for any clarification!
Warm regards,
Andrew
Hi,
Thanks for your question! loss is the total loss as described in the paper, which is the sum of mixmatch_loss+aet_loss+kl_loss. Here Closs is the cross-entropy loss in MixMatch loss. Eloss is another term in MixMatch loss, which is a MSE loss as described in MixMatch paper. wd is the weight for the Eloss. waet is the weight for aet-loss1(projective transformation), other weights for aet loss also follow the same trend to warm up. top1avg is the accuracy when I used the average predictions of all different transformation.
Any questions, welcome to update in this issue. If your question is resolved, please close the issue. Thanks!