Data containing empty or nan

[anthonysirico]

Here is what the data looks like:

The first column "Adjacency Matrix" is to create the graphs (obviously), the other 5 columns are graph-level features.

1. The NaN and empty cells is significant to the data, it tell us that the graph did not simulate. Can PyG/PyTorch read empty or NaN?
2. If yes to both, which is better NaN or empty?
3. If not, any suggestions to replace the value?

[anthonysirico]

#7461

[rusty1s]

Yes, PyTorch can handle NaNs in tensor data. However, gradients will be NaN in this case, so you need a way to impute it before using it as input into a neural network.

[anthonysirico]

thanks, I ended up using torch.nan_to_num