Add additional datasets

[cthoyt]

Summary

This PR adds the OncoPolyPharmacology, adds a new dataloader that does all processing locally, and updates the pipeline for continuous labels (e.g., for drug synergy) as opposed to binary labels.

• [x] Code passes all tests
• [x] Unit tests provided for these changes
• [x] Documentation and docstrings added for these changes
• [x] Add other datasets from https://github.com/AstraZeneca/chemicalx/issues/3#issuecomment-996231922

Changes

• Add new harness for local preprocessing and storage of datasets
• Add mock data for local dataset loader and associated tests
• Add OncoPolyPharmacology
• Enable selection of evaluation metrics. ROC-AUC is not appropriate for drug synergy tasks, where the values are continuous, so MSE and MAE are given as an example.

[codecov-commenter]

Codecov Report

Merging #61 (7acb34c) into main (30d3327) will decrease coverage by 6.23%. The diff coverage is 52.33%.

@@            Coverage Diff             @@
##             main      #61      +/-   ##
==========================================
- Coverage   98.07%   91.83%   -6.24%     
==========================================
  Files          28       29       +1     
  Lines         675      772      +97     
==========================================
+ Hits          662      709      +47     
- Misses         13       63      +50     

Impacted Files	Coverage Δ
chemicalx/data/drugfeatureset.py	100.00% <ø> (ø)
chemicalx/data/datasetloader.py	77.34% <44.44%> (-17.90%)	:arrow_down:
chemicalx/data/utils.py	45.23% <45.23%> (ø)
chemicalx/data/contextfeatureset.py	90.00% <75.00%> (-10.00%)	:arrow_down:
chemicalx/pipeline.py	88.57% <84.61%> (-0.14%)	:arrow_down:
chemicalx/data/__init__.py	100.00% <100.00%> (ø)
tests/unit/test_models.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 30d3327...7acb34c. Read the comment docs.

[cthoyt]

@benedekrozemberczki I could do a mock for the OncoPolyPharmacology dataset, but maybe it makes more sense to apply the loader class to a mock dataset instead

Edit: done in d8309f4