Trying to train MatchNet but getting issues output loss calculation.

[mukunda11]

MatchNet_output_error.txt
MatchNet_output.txt

[mukunda11]

I see that someone else had this issue before. Unfortunately, I will not be able to redownload pytorch with pip because I do not have sudo access to the server. I hope there is another way to solve the problem.

[Alfredo Canziani]

So, the error you're getting is

File "/depot/chubykin/gpu/pytorch-CortexNet-master/model/Model02.py", line 76, in forward
    s = state[layer - 1] or V(x.data.clone().zero_())
    torch.typename(self.data) + " is ambiguous")
RuntimeError: bool value of Variable objects containing non-empty torch.cuda.FloatTensor is ambiguous

where it used to be legal to check if an object was not None.
Now you have to rewrite

s = state[layer - 1] or V(x.data.clone().zero_())

as

s = state[layer - 1] if state[layer - 1] is not None else V(x.data.clone().zero_())

This should solve your issue.
If it does, you're more then welcome to send a PR fixing this and similar deprecated syntaxes.

If my reply fixes your problem, you can close this issue.

[mukunda11]

Thank you! That fixed the error. However, I am getting the same error for another line.
Is there a similar issue for this function in the main.py file?

[Alfredo Canziani]

Please, don't upload pictures... Just type code and enclose it in three back-ticks ```. You can add the language too by typing python after the ticks.

Anyhow, yes, replace if not h with if h is None.

Also, if you installed the latest PyTorch (version 0.4.0), then return V(h.data) should be replaced by return h.detach().

[mukunda11]

Apologies for the picture. Thank you for the quick response.