[ENH] Cython axis-aligned multi-view splitter

[adam2392]

Closes #92

Changes proposed in this pull request:

• Adds an experimental Cython API for multi-view splitting

I foresee multi-view as a separate submodule for now since it's a rather experimental feature that is there to support MIGHT/Co-MIGHT in Cancer research.

Multiview splitting can analogously have its axis-aligned, oblique and morf versions imo. This initial PR implements the axis-aligned multi-view approach.

Note: what is not done is tracking constants, which is an important feature.

Before submitting

• I've read and followed all steps in the Making a pull request
  section of the CONTRIBUTING docs.
• I've updated or added any relevant docstrings following the syntax described in the
  Writing docstrings section of the CONTRIBUTING docs.
• If this PR fixes a bug, I've added a test that will fail without my fix.
• If this PR adds a new feature, I've added tests that sufficiently cover my new functionality.

After submitting

• All GitHub Actions jobs for my pull request have passed.

[adam2392]

Possible code to sample different feature sets:

        # dataset1 from 0 to N
        for i in range(max_features//2):
            index = random_index_selection(dataset_1)

            rand_int(0, N, random_state)

        # dataset2 from N+1 to M 
        for i in range(max_features//2):
            index = random_index_selection(dataset_2)

            rand_int(N+1, M, random_state)

[adam2392]

Mainly files to review:

• sktree/tree/_multiview.py
• sktree/tree/_oblique_splitter.pxd
• sktree/tree/_oblique_splitter.pyx
• examples/splitters/plot_multiview_axis_aligned_splitter.py

[adam2392]

https://output.circle-artifacts.com/output/job/56f3cf6c-74a2-437c-8bc7-e86da9e52d6f/artifacts/0/dev/auto_examples/splitters/plot_multiview_axis_aligned_splitter.html#sphx-glr-auto-examples-splitters-plot-multiview-axis-aligned-splitter-py

[sampan501]

Looks good to me! @adam2392 will there be another PR exposing a class for the multiview hypothesis test? Maybe in forestht.py?

[adam2392]

Looks good to me! @adam2392 will there be another PR exposing a class for the multiview hypothesis test? Maybe in forestht.py?

The design allows any arbitrary tree to be used in HonestForest, and the hypothesis test allows any arbitrary HonestForest parametrization to be used, so in principle this is "possible" once this PR is merged.

However, the next PR will show a more explicit example of multi-view learning on simulated data in examples/ to test the API out. I think there should be some guardrails in place on checking values of feature_set_ends relative to X. I need to add these still tho.

[sampan501]

I agree, I just think it would be easier for the user if the whole class was exposed the user. I imagine a class kind of like FeatureImportanceForestClassifier that just calls HonestForest with the correct parameters. I guess I'm referring more to the Multiview test and not specifically the splitter

[codecov]

Codecov Report

Attention: 4 lines in your changes are missing coverage. Please review.

Comparison is base (6760662) 88.78% compared to head (e2ee6e5) 89.76%.
Report is 7 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #129      +/-   ##
==========================================
+ Coverage   88.78%   89.76%   +0.98%     
==========================================
  Files          37       41       +4     
  Lines        3138     3351     +213     
==========================================
+ Hits         2786     3008     +222     
+ Misses        352      343       -9     

Files	Coverage Δ
sktree/__init__.py	80.76% <100.00%> (ø)
sktree/ensemble/__init__.py	100.00% <100.00%> (ø)
sktree/ensemble/_multiview.py	100.00% <100.00%> (ø)
sktree/experimental/sdf.py	69.23% <ø> (ø)
sktree/stats/tests/test_coleman.py	100.00% <ø> (ø)
sktree/stats/utils.py	91.89% <ø> (ø)
sktree/tests/test_multiview_forest.py	100.00% <100.00%> (ø)
sktree/tests/test_supervised_forest.py	99.41% <100.00%> (+<0.01%)	⬆️
sktree/tree/__init__.py	100.00% <ø> (ø)
sktree/tree/_classes.py	75.00% <ø> (ø)
... and 6 more

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[adam2392]

I agree, I just think it would be easier for the user if the whole class was exposed the user. I imagine a class kind of like FeatureImportanceForestClassifier that just calls HonestForest with the correct parameters. I guess I'm referring more to the Multiview test and not specifically the splitter

Sure yeah I agree it'll be easier if we can expose a sensible interface that makes semantic sense for users. We can play around w/ the best way to do this.