Add unbatch functionality

[tzuhanchang]

Unbatching data from DataLoader batch to a list. This is useful for GNN edge classifier, in which case, graphs are required to be reconstructed with edge predictions.

[codecov]

Codecov Report

Merging #4628 (7365414) into master (da78713) will increase coverage by 0.00%.
The diff coverage is 100.00%.

❗ Current head 7365414 differs from pull request most recent head 3992f87. Consider uploading reports for the commit 3992f87 to get more accurate results

@@           Coverage Diff           @@
##           master    #4628   +/-   ##
=======================================
  Coverage   82.97%   82.98%           
=======================================
  Files         316      317    +1     
  Lines       16784    16792    +8     
=======================================
+ Hits        13927    13935    +8     
  Misses       2857     2857           

Impacted Files	Coverage Δ
torch_geometric/utils/__init__.py	100.00% <100.00%> (ø)
torch_geometric/utils/unbatch.py	100.00% <100.00%> (ø)

---

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 da78713...3992f87. Read the comment docs.

[rusty1s]

I think in order to get this merged, we would need to have a pure PyTorch function without for-loops. IMO, this is achievable by

src.split(degree(index).tolist())

In addition, we would need to add this function to utils rather than loader/utils.py, add doccumentation, and write tests for it. WDYT?

[tzuhanchang]

@rusty1s Thank you very much! src.split(degree(index).tolist()) works perfectly, it can replace two for loops completely. I also add it to utils.

[rusty1s]

This looks better. Thank you! Can we also add some basic tests in test/utils/test_unbatch.py and add this operator to torch_geometric/utils/__init__.py?

[rusty1s]

Thanks for the updates!