Fix R EBM outer bag score aggregation

[TemidayoA]

Summary

• Accumulate term scores from every outer bag before averaging them.
• Add an R regression test for imbalanced binary classification.
• Include the R tests directory in the package build.

Root cause

ebm_classify() overwrote term_scores during each outer-bag iteration and then divided only the final bag’s scores by outer_bags. As the bag count increased, this shrank the logits toward zero and pushed predicted probabilities toward 0.5.

The problem was therefore incorrect outer-bag aggregation, not a missing intercept.

Using the dataset from the issue, the mean predicted probability with 16 bags changed from 0.390069 to 0.100096, closely matching the observed positive rate of 0.1.

Validation

• R CMD check --no-manual passed, with one existing installed-size note.
• The regression test passes for the original imbalanced dataset.
• Mean predictions remain consistent with outer_bags set to 1, 2, 4, 8, and 16.

Fixes #417

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 67.20%. Comparing base (2ed5ecb) to head (5273772).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #677      +/-   ##
==========================================
- Coverage   67.25%   67.20%   -0.06%     
==========================================
  Files          77       77              
  Lines       11735    11735              
==========================================
- Hits         7892     7886       -6     
- Misses       3843     3849       +6     

Flag	Coverage Δ
bdist_linux_311_python	66.97% <ø> (ø)
bdist_linux_312_python	66.95% <ø> (-0.02%)	⬇️
bdist_linux_313_python	66.97% <ø> (ø)
bdist_linux_314_python	66.87% <ø> (ø)
bdist_linuxarm_311_python	66.98% <ø> (ø)
bdist_linuxarm_312_python	66.98% <ø> (ø)
bdist_linuxarm_313_python	66.98% <ø> (ø)
bdist_linuxarm_314_python	66.87% <ø> (-0.04%)	⬇️
bdist_mac_311_python	67.11% <ø> (-0.04%)	⬇️
bdist_mac_312_python	67.13% <ø> (-0.02%)	⬇️
bdist_mac_313_python	67.11% <ø> (-0.02%)	⬇️
bdist_mac_314_python	67.03% <ø> (ø)
bdist_win_311_python	67.14% <ø> (-0.04%)	⬇️
bdist_win_312_python	67.13% <ø> (-0.04%)	⬇️
bdist_win_313_python	67.13% <ø> (-0.06%)	⬇️
bdist_win_314_python	67.03% <ø> (-0.04%)	⬇️
sdist_linux_311_python	?
sdist_linux_312_python	?
sdist_linux_313_python	?
sdist_linux_314_python	?
sdist_linuxarm_311_python	?
sdist_linuxarm_312_python	?
sdist_linuxarm_313_python	?
sdist_linuxarm_314_python	?
sdist_mac_311_python	?
sdist_mac_312_python	?
sdist_mac_313_python	?
sdist_mac_314_python	?
sdist_win_311_python	?
sdist_win_312_python	?
sdist_win_313_python	?
sdist_win_314_python	?

Flags with carried forward coverage won't be shown. Click here to find out more.

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[TemidayoA]

CI note: I investigated the three failing checks in run 31942930664. They are unrelated to this R-only change:

• format_ruff encounters a 403: Resource not accessible by integration while its reporting step attempts to create a check-run from the fork PR.
• docs reports execution failures across multiple documentation notebooks, followed by the same check-run permission error. This PR does not modify any documentation or Python files.
• testS (win_314_python) cannot build interpret-core because MSBuild.exe is unavailable on the runner; the subsequent test step also reports that pytest is unavailable.

The change was validated locally with R CMD check --no-manual, the new regression test, and the original imbalanced-data reproduction across outer_bags values 1, 2, 4, 8, and 16.

Could a maintainer please rerun the affected checks or confirm whether these are known workflow issues?

[paulbkoch]

Thanks @TemidayoA! I agree the failing tests are unrelated to this PR. The testR CI step for the R package completed without error, so I will merge. Good find!

[TemidayoA]

Thank you so much, @paulbkoch! I really appreciate you reviewing the PR, confirming the unrelated checks, and merging the fix. Glad I could help track this down, looking forward to contributing more