Added PyTorch Sparse Tensor support for remove_self_loops, add_self_loops, and maybe_num_nodes

[EdisonLeeeee]

This PR aims to add PyTorch Sparse Tensor support for remove_self_loops, add_self_loops, and maybe_num_nodes, which will be used for supporting PyTorch Sparse Tensor in torch_geometric.nn.conv, e.g. AGNNConv:

pytorch_geometric/torch_geometric/nn/conv/agnn_conv.py

Lines 65 to 69 in 09dbb20

	if self.add_self_loops:
	if isinstance(edge_index, Tensor):
	edge_index, _ = remove_self_loops(edge_index)
	edge_index, _ = add_self_loops(edge_index,
	num_nodes=x.size(self.node_dim))

There are some questions to be clarified:

• Should I also add support for other functions like add_remaining_self_loops?
• Should I also update the doc-strings correspondingly?

[codecov]

Codecov Report

Merging #6847 (935be98) into master (5247917) will decrease coverage by 0.24%.
The diff coverage is 100.00%.

❗ Current head 935be98 differs from pull request most recent head 985bf6b. Consider uploading reports for the commit 985bf6b to get more accurate results

@@            Coverage Diff             @@
##           master    #6847      +/-   ##
==========================================
- Coverage   91.74%   91.50%   -0.24%     
==========================================
  Files         427      427              
  Lines       23173    23076      -97     
==========================================
- Hits        21259    21116     -143     
- Misses       1914     1960      +46     

Impacted Files	Coverage Δ
torch_geometric/utils/loop.py	100.00% <100.00%> (ø)
torch_geometric/utils/num_nodes.py	96.77% <100.00%> (+0.34%)	⬆️
torch_geometric/visualization/graph.py	68.25% <0.00%> (-26.99%)	⬇️
torch_geometric/nn/conv/utils/typing.py	83.75% <0.00%> (-15.00%)	⬇️
torch_geometric/nn/pool/asap.py	91.89% <0.00%> (-8.11%)	⬇️
torch_geometric/nn/models/attentive_fp.py	95.83% <0.00%> (-4.17%)	⬇️
torch_geometric/io/tu.py	90.80% <0.00%> (-2.30%)	⬇️
torch_geometric/transforms/virtual_node.py	93.47% <0.00%> (-2.18%)	⬇️
torch_geometric/nn/models/basic_gnn.py	89.26% <0.00%> (-1.70%)	⬇️
torch_geometric/nn/glob.py	76.92% <0.00%> (-1.65%)	⬇️
... and 30 more

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[wsad1]

to_troch_coo_tensor returns a sparse torch.Tensor. It would be nice to add a test for SparseTensor too.

[EdisonLeeeee]

Do you mean to add tests with torch_sparse.SparseTensor as input for these functions as well?

[wsad1]

Actually skip that. I guess we only want to support sparse torch.Tensor here, right?

[EdisonLeeeee]

Yes, exactly.