Unsupervised Heterogeneous Graph Learning

[Yonggie]

added unsupervised heterogeneous graph representation learning example.

[codecov-commenter]

Codecov Report

Merging #3189 (e9d614f) into master (0847344) will decrease coverage by 0.12%.
The diff coverage is n/a.

❗ Current head e9d614f differs from pull request most recent head 690f14b. Consider uploading reports for the commit 690f14b to get more accurate results

@@            Coverage Diff             @@
##           master    #3189      +/-   ##
==========================================
- Coverage   82.44%   82.31%   -0.13%     
==========================================
  Files         310      310              
  Lines       15934    15932       -2     
==========================================
- Hits        13136    13115      -21     
- Misses       2798     2817      +19     

Impacted Files	Coverage Δ
torch_geometric/nn/conv/utils/typing.py	81.25% <0.00%> (-17.50%)	⬇️
torch_geometric/io/tu.py	93.58% <0.00%> (-2.57%)	⬇️
torch_geometric/nn/models/mlp.py	98.41% <0.00%> (-1.59%)	⬇️
torch_geometric/transforms/gdc.py	77.94% <0.00%> (-1.03%)	⬇️
torch_geometric/data/dataset.py	96.80% <0.00%> (-0.80%)	⬇️
torch_geometric/nn/conv/rgat_conv.py	83.76% <0.00%> (-0.53%)	⬇️
torch_geometric/graphgym/utils/comp_budget.py	15.51% <0.00%> (+0.51%)	⬆️

---

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 0847344...690f14b. Read the comment docs.

[Yonggie]

@rusty1s Hi, it's ready for review :)

[vincentvic]

Use the class AddMetaPaths for a cleaner implementation

[Yonggie]

It's been quite a long time since I wrote this code. I may take a while to catch it up again.

[Yonggie]

I implemented DMGI using torch_geometric, but it seems the performance is not as good as the paper said. Original implementation: https://github.com/pcy1302/DMGI

[rusty1s]

Thanks for the updates @Yonggie, and sorry for the delay. We will try to get this merged in the upcoming week.

[rusty1s]

Do you know of any differences between this implementation and the official one?

[Yonggie]

I just found a small difference in the encoder and changed the code. but it still seems not good enough. And another weird thing is valid score and test score are exactly the same.

[rusty1s]

Thank you :)