Open richardrl opened 3 years ago
That is intended. All tensor data should be saved as PyTorch tensors.
@rusty1s Alright, but it could be helpful to have an assertion or check for whether the input is a torch tensor. it was tricky to trace this error
You are right. I will add that in.
Ostensibly, we now have a valid data object. But none of the subsequent operations over this object are supported when there are numpy arrays.
For example, in line 204 of torch_geometric/data/data.py, there is a line:
item.size(...)
This will fail if item is a numpy array, and succeed if item is a torch tensor.