Loading Custom Data with Graph Level Features

[michaelweihaosong]

Hi all, I am new to loading custom data and have been struggling with this issue. I am on torch_geometric v1.7.2

Specifically, I have created a torch_geometric.data.Dataset with __getitem__ and __len__ implemented.

The __getitem__ returns data, zrirad, labels, data_decoy, where data is a torch_geometric.data.Data(x, edge_index), zrirad is a (1,15) tensor graph level features, labels is a (1,3) tensor graph level labels, and data_decoy is just a string.

Below I pasted the error msgs. It seems to be some collate_fn related issue but I couldn't figure it out. Any help is welcomed, thanks very much!

for _, (data, zrirad, labels, data_decoy) in enumerate(train_bar):
  File "/home/mikesmac/anaconda3/envs/DeepDock/lib/python3.7/site-packages/tqdm/std.py", line 1185, in __iter__
    for obj in iterable:
  File "/home/mikesmac/anaconda3/envs/DeepDock/lib/python3.7/site-packages/torch/utils/data/dataloader.py", line 435, in __next__
    data = self._next_data()
  File "/home/mikesmac/anaconda3/envs/DeepDock/lib/python3.7/site-packages/torch/utils/data/dataloader.py", line 475, in _next_data
    data = self._dataset_fetcher.fetch(index)  # may raise StopIteration
  File "/home/mikesmac/anaconda3/envs/DeepDock/lib/python3.7/site-packages/torch/utils/data/_utils/fetch.py", line 47, in fetch
    return self.collate_fn(data)
  File "/home/mikesmac/anaconda3/envs/DeepDock/lib/python3.7/site-packages/torch_geometric/data/dataloader.py", line 35, in __call__
    return self.collate(batch)
  File "/home/mikesmac/anaconda3/envs/DeepDock/lib/python3.7/site-packages/torch_geometric/data/dataloader.py", line 30, in collate
    return [self.collate(s) for s in zip(*batch)]
  File "/home/mikesmac/anaconda3/envs/DeepDock/lib/python3.7/site-packages/torch_geometric/data/dataloader.py", line 30, in <listcomp>
    return [self.collate(s) for s in zip(*batch)]
  File "/home/mikesmac/anaconda3/envs/DeepDock/lib/python3.7/site-packages/torch_geometric/data/dataloader.py", line 18, in collate
    self.exclude_keys)
  File "/home/mikesmac/anaconda3/envs/DeepDock/lib/python3.7/site-packages/torch_geometric/data/batch.py", line 110, in from_data_list
    inc = data.__inc__(key, item)
  File "/home/mikesmac/anaconda3/envs/DeepDock/lib/python3.7/site-packages/torch_geometric/data/data.py", line 190, in __inc__
    return self.num_nodes if bool(re.search('(index|face)', key)) else 0
  File "/home/mikesmac/anaconda3/envs/DeepDock/lib/python3.7/site-packages/torch_geometric/data/data.py", line 215, in num_nodes
    return item.size(self.__cat_dim__(key, item))
TypeError: 'int' object is not callable

[rusty1s]

It looks like data.x is an integer, not a PyTorch tensor. Can you confirm?

[michaelweihaosong]

Yes you are right about this, it is a np.array. I have casted it as a pytorch tensor and the code works now! Thank you!

One question though, does the mini-batching in torch_geometric handle tensors of all data types such as string? For example, if my getitem returns Data, a string tensor, some float tensors, etc.

[rusty1s]

Yes, that should be supported. For any item that is not a PyG data object, we use the default PyTorch collate_fn.