inconsistent tensor size about pygcn HOT 6 CLOSED

Interesting, thanks for reporting. Did you perform any changes to the code? What steps can I take to reproduce this error?

from pygcn.

I changed line 46 in train.py from
model = GCN(nfeat=features.shape[1],
to
model = GCN(nfeat=features.size()[1],

otherwise there will be an error message says tensorflow object does not have 'shape'.

That's the only thing I changed.

from pygcn.

Might be a PyTorch version issue. I only tested this with 0.2.0. Are you on 0.3.0?

from pygcn.

I checked with PyTorch 0.3.0 (latest release version) and it still works fine. features.shape[1] is a supported operation in my environment. Can you check which version you are on?

from pygcn.

I use 0.1.12_2
I think I need to upgrade it.
Thank you very much for your reply.

from pygcn.

Glad we found the issue. I would certainly recommend upgrading.

from pygcn.