Add reject-option-classification explainer - #565
Merged
yakew7 merged 1 commit intoSep 10, 2026
Conversation
Adds explainers/reject-option-classification.md covering Reject Option Classification (Kamiran, Karim & Zhang, IEEE ICDM 2012), the post-processing method that reassigns labels only inside a low-confidence band around the decision boundary. - Explicit name-collision note distinguishing it from reject-inference.md and a contrast table against this repo's ThresholdOptimizer (S4). - Worked example on the COMPAS baseline logistic regression: sweeps the critical-region half-width theta and shows the fairness gap shrink at +-0.10 (651 of 3,254 predictions flipped, gap halved), then reverse to -70 pp and collapse accuracy at +-0.15. - Detection code: reject_option_classify() plus a sweep_theta() helper. - Limitations: theta has no principled default, needs group membership at prediction time, can invert the disparity, and the individual-fairness objection. Regenerated assets/explainers-data.js, faircode/_explainers/, the HTML page, OG images, sitemap.xml and llms-full.txt; check_explainer_count.py --fix bumped the 60 -> 61 mentions in README/CONTRIBUTORS/METRICS/ROADMAP.
propcgamer20-png
requested review from
Shreyash0712 and
yakew7
as code owners
September 10, 2026 09:33
Contributor
|
@propcgamer20-png is attempting to deploy a commit to the yashkewlani2020-gmailcom's projects Team on Vercel. A member of the Team first needs to authorize it. |
|
@yakew7 @Shreyash0712 - new PR touching a path you own, please review. |
yakew7
added a commit
that referenced
this pull request
Sep 10, 2026
ROADMAP.md's "Current traction" table still said 60 explainers (contradicting its own Phase 1 status line, already correctly saying 61) and llms.txt was missing the reject-option-classification entry, both because PR #565 added the 61st explainer after PR #562 fixed this same drift. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
yakew7
added a commit
that referenced
this pull request
Sep 10, 2026
Documents the newest explainer (#565), the CODEOWNERS grant to @propcgamer20-png, 8 doc/code fixes (#557-564, #566, #567), and the two direct fixes made during this review: the README.md regression from #557's broad substring replacement, and the recurred llms.txt/ROADMAP.md explainer-count drift after #565. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes #542.
What
New explainer
explainers/reject-option-classification.mdon Reject Option Classification (Kamiran, Karim & Zhang, Decision Theory for Discrimination-Aware Classification, IEEE ICDM 2012, pp. 924-929). ROC is a post-processing method: it leaves the model untouched and only reassigns labels for instances that fall inside a low-confidence band of half-width theta around the decision boundary, pushing them toward the disadvantaged group.Notes for review
reject-inference.md(missing-outcome correction in credit scoring). Different fields, different problems.ThresholdOptimizerso a reader knows when ROC is and isn't the tool already available here.sweep_theta()helper walks the half-width: the fairness gap barely moves at +-0.05, roughly halves at +-0.10 (651 of 3,254 test predictions flipped), then at +-0.15 overcorrects to about -70 pp and drops accuracy to ~45%. The point being that theta has no principled default and the method can invert the disparity it was meant to close.reject_option_classify()andsweep_theta().Generated files
Ran
scripts/build_explainers.pyandscripts/check_explainer_count.py --fix. Regenerated:assets/explainers-data.js,faircode/_explainers/{data.json,reject-option-classification.md},explainers/reject-option-classification.html,assets/og{,-light}/reject-option-classification.png, one newsitemap.xmlentry, and the appended section inllms-full.txt. The 60 -> 61 count bump touchesREADME.md,CONTRIBUTORS.md,METRICS.md,ROADMAP.md.check_em_dash.py,check_explainer_count.py,check_broken_links.py,check_generated_files_current.pyandtests/test_build_explainers.pyall pass.