graytowne / revisit_adv_rec Goto Github PK

I find some questions in the function of evaluate_epoch in the adv_trainer.py, which lose the function trainer.validation.Could you help me fix the problem?

Excuse me, when I used the ncf model or even a simple mlp as the surrogate model, it always reported the error of "One of the differentiated Tensors appears to not have been used in the graph. Set allow_unused=True if this is the desired behavior." when the program ran to this line: "adv_grads = torch.autograd.grad(adv_loss, data_tensor)[0]" in fit_adv() function. However, when I added the "allow_unused=True", the adv_grads returned is None. I was puzzled for a long time. Please help me to fix the problem. orz.

When the surrogate model is "sur_wmf_afs", the program reports the error of "RuntimeError: a view of a leaf Variable that requires grad is being used in an in-place operation."