Closed cothurn closed 5 years ago
which version of pytorch are you using?
0.4.0
I was using 0.3.1. Can you quickly try with this version?
I'll look into 0.4.0.
0.3.1 Works.
I wanted to use the bottleneck functionality of 0.4.0, so I am testing with that version. The above error can be avoided by manually converting the tensors to cuda tensors(which leads to the same type error in other locations). In CPU mode, however, I receive this error:
Traceback (most recent call last):
File "main.py", line 142, in <module>
avg_loss = loop_dataset(cooked_data_dict['train'], regressor, train_idxes, optimizer, start_iter=i, n_iters=valid_interval)
File "main.py", line 64, in loop_dataset
mae = mae.data.numpy()[0]
IndexError: too many indices for array
mae is a scalar in 0.4.0. If you want to keep with 0.4.0, try mae = mae.data.item()
This is what I get running main.py:
The issue here is
RuntimeError: Expected object of type torch.cuda.sparse.FloatTensor but found type torch.sparse.FloatTensor
Not sure the cause of this, as searching torch.cuda.sparse.FloatTensor on google returns nothign revelant.