[Bug] IndexError: index 0 is out of bounds for dimension 0 with size 0

[bhadreshpsavani]

In the retrosynthesis tutorial

from torchdrug.utils import plot

for i in range(2):
    sample = reaction_dataset[i]
    reactant, product = sample["graph"]
    reactants = reactant.connected_components()[0]
    products = product.connected_components()[0]
    plot.reaction(reactants, products)

gives below error

---------------------------------------------------------------------------
IndexError                                Traceback (most recent call last)
Input In [14], in <cell line: 5>()
      8 reactants = reactant.connected_components()[0]
      9 products = product.connected_components()[0]
---> 10 plot.reaction(reactants, products)

File /usr/local/lib/python3.9/dist-packages/torchdrug/utils/plot.py:26, in reaction(reactants, products, save_file, figure_size, atom_map)
     14 """
     15 Visualize a chemical reaction.
     16 
   (...)
     23     atom_map (bool, optional): visualize atom mapping or not
     24 """
     25 rxn = AllChem.ChemicalReaction()
---> 26 for reactant in reactants:
     27     mol = reactant.to_molecule()
     28     if not atom_map:

File /usr/local/lib/python3.9/dist-packages/torchdrug/data/graph.py:1159, in PackedGraph.__next__(self)
   1157 def __next__(self):
   1158     if self._iter_index < self.batch_size:
-> 1159         item = self[self._iter_index]
   1160         self._iter_index += 1
   1161         return item

File /usr/local/lib/python3.9/dist-packages/torchdrug/data/graph.py:1458, in PackedGraph.__getitem__(self, index)
   1455     index = (index,)
   1457 if isinstance(index[0], int):
-> 1458     item = self.get_item(index[0])
   1459     if len(index) > 1:
   1460         item = item[index[1:]]

File /usr/local/lib/python3.9/dist-packages/torchdrug/data/graph.py:1378, in PackedGraph.get_item(self, index)
   1376 edge_list = self.edge_list[edge_index].clone()
   1377 edge_list[:, :2] -= self._offsets[edge_index].unsqueeze(-1)
-> 1378 data_dict, meta_dict = self.data_mask(node_index, edge_index, graph_index=graph_index)
   1380 return self.unpacked_type(edge_list, edge_weight=self.edge_weight[edge_index], num_node=self.num_nodes[index],
   1381                           num_relation=self.num_relation, meta_dict=meta_dict, **data_dict)

File /usr/local/lib/python3.9/dist-packages/torchdrug/data/graph.py:1433, in PackedGraph.data_mask(self, node_index, edge_index, graph_index, include, exclude)
   1431     v = v[edge_index]
   1432 elif type == "graph" and graph_index is not None:
-> 1433     v = v[graph_index]
   1434 elif type == "node reference" and node_index is not None:
   1435     if node_mapping is None:

IndexError: index 0 is out of bounds for dimension 0 with size 0

[Oxer11]

Hi! Thanks for pointing out this issue. This bug is caused by a typo in the commit https://github.com/DeepGraphLearning/torchdrug/commit/f8d2de6fc91c1ff60a62b0279d61d1be8e1638ff, which slightly modified the data.Graph.split() method. I've fixed this typo in the commit https://github.com/DeepGraphLearning/torchdrug/commit/5e07dd36b6f7b58532315cbd99fd2e015d12d808 and it should work now.

[bhadreshpsavani]

Thanks, @KiddoZhu, This issue is resolved with the latest commit