Add LinkNeighborLoader to Pytorch Lightning datamodule

[ar104]

Extend the pytorch lightning datamodule to support wrapping the link neighbor loader.

[codecov]

Codecov Report

Merging #4868 (3e7c7e2) into master (65ab1e0) will decrease coverage by 0.13%.
The diff coverage is 26.66%.

@@            Coverage Diff             @@
##           master    #4868      +/-   ##
==========================================
- Coverage   82.76%   82.62%   -0.14%     
==========================================
  Files         329      329              
  Lines       17709    17748      +39     
==========================================
+ Hits        14656    14665       +9     
- Misses       3053     3083      +30     

Impacted Files	Coverage Δ
torch_geometric/data/lightning_datamodule.py	33.75% <23.25%> (-3.75%)	⬇️
torch_geometric/data/__init__.py	100.00% <100.00%> (ø)
torch_geometric/loader/link_neighbor_loader.py	92.53% <100.00%> (+0.68%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 65ab1e0...3e7c7e2. Read the comment docs.

[rusty1s]

@ar104 I think this generally looks good although I would prefer to move this functionality to its own module, e.g., LightningLinkData. Otherwise, I feel the interface of input_nodes might become a bit confusing to you. WDYT?

[ar104]

@ar104 I think this generally looks good although I would prefer to move this functionality to its own module, e.g., LightningLinkData. Otherwise, I feel the interface of input_nodes might become a bit confusing to you. WDYT?

Done.

[wsad1]

It's unclear what would happen when train/test/val edges are None. If input_val_edges or input_test_edges are None then all edges would be considered.

[rusty1s]

+1. I would be okay to make them required for now, and add a TODO to revisit this later on.

[rusty1s]

+1. I would be okay to make them required for now, and add a TODO to revisit this later on.