First commit for heterogeneous graph support

[doGregor]

This includes:

• a new data-structure StaticHeteroGraphTemporalSignal that works like StaticGraphTemporalSignal but with torch_geometric HeteroData objects as snapshots; instead of np arrays, dictionaries wirth key(node/edge types as strings)/value(indices/features as np arrays) are expected
• code tests for the new data-structure
• documentation for the new data-structure

Feedback is welcome! If you think it makes sense to extend your package like that it would be great to open a new branch for heterogeneous support where we can go on working in this direction.
Thanks and cheers, Gregor

[benedekrozemberczki]

Hi there @doGregor , it seems to depend on a version of PyG that already has the hetero data module. When was that introduced? Which version?

[doGregor]

Hi @benedekrozemberczki , heterogeneous graphs have been introduced in PyG version 2.0.0 that was released on september 13 2021

[benedekrozemberczki]

The 0.5.0 support PyG 2.0 please update the PR and let us go from there @doGregor !

[doGregor]

Great, thank you! Updated it.

[codecov-commenter]

Codecov Report

Merging #125 (62a86b4) into master (5d5c7dd) will decrease coverage by 1.04%.
The diff coverage is 92.51%.

❗ Current head 62a86b4 differs from pull request most recent head 0437310. Consider uploading reports for the commit 0437310 to get more accurate results

@@            Coverage Diff             @@
##           master     #125      +/-   ##
==========================================
- Coverage   99.16%   98.11%   -1.05%     
==========================================
  Files          48       50       +2     
  Lines        3829     4041     +212     
==========================================
+ Hits         3797     3965     +168     
- Misses         32       76      +44     

Impacted Files	Coverage Δ
...ignal/static_hetero_graph_temporal_signal_batch.py	89.58% <89.58%> (ø)
...oral/signal/static_hetero_graph_temporal_signal.py	91.86% <91.86%> (ø)
test/batch_test.py	100.00% <100.00%> (ø)
test/dataset_test.py	94.15% <100.00%> (-5.85%)	⬇️
torch_geometric_temporal/signal/__init__.py	100.00% <100.00%> (ø)
...orch_geometric_temporal/signal/train_test_split.py	93.10% <0.00%> (-6.90%)	⬇️
...ic_temporal/signal/static_graph_temporal_signal.py	97.26% <0.00%> (-2.74%)	⬇️
...c_temporal/signal/dynamic_graph_temporal_signal.py	97.33% <0.00%> (-2.67%)	⬇️
...ric_temporal/signal/dynamic_graph_static_signal.py	94.73% <0.00%> (-2.64%)	⬇️

---

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 5d5c7dd...0437310. Read the comment docs.

[benedekrozemberczki]

I would suggest increasing the test coverage first.

[benedekrozemberczki]

Is this ready for review @doGregor ?

[doGregor]

Hi @benedekrozemberczki , yes, the new data structures are ready to review. What's left afterwards is updating the train/test split method and finally the layers. Will go on with that soon...

[benedekrozemberczki]

These are really cool things.

[benedekrozemberczki]

Merged these for now!