Explainer precision recall curve - #275
Merged
Merged
Conversation
Contributor
|
@evanjain-dot 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. |
Owner
|
@Shreyash0712 can you review this? |
yakew7
added a commit
that referenced
this pull request
Aug 17, 2026
img.save(..., optimize=True) in scripts/generate_og_images.py runs a per-scanline filter/strategy search whose "best" pick can differ by a few bytes between zlib builds - same pixel content, different platform, different output bytes. That's why a macOS-generated commit could never satisfy CI's Ubuntu-side fresh regeneration in build-explainers.yml's byte-exact drift check, no matter how many times it was regenerated and recommitted (this is what broke again after #262/#275, the same failure mode as before). Switched to compress_level=9: a single deterministic deflate pass with no heuristic search. Verified stable, identical output across repeated local runs. Regenerated all 47 OG images plus the two real drift lines (precision-recall-curve.html, sitemap.xml) and fixed a malformed markdown line in CHANGELOG.md's #92 entry left over from an earlier edit.
Contributor
|
Welp just when I thought I'll check it out. Don't see any issues on a glance, but I did notice the third point in Further Reading resource isn't linked. |
Owner
|
@Shreyash0712 timing is crazy you mind opening a pr to fix the linking issue in it and other explainers asw if u find it? |
Owner
|
If you do make a pr js make sure to run the generation of images or the ci fails |
Contributor
|
okay, will do |
yakew7
added a commit
that referenced
this pull request
Aug 17, 2026
Two new merges since the last snapshot (#266): - #275 (evanjain-dot): the precision-recall-curve explainer (#92). 6 merged PRs, 28 commits now - stays in first place, further ahead. - #267 (Circout-sudo): a small Makefile refinement to the coverage target (comment clarity + a -q flag). 3 merged PRs, 3 commits now - moves their entry up from the tied-at-2 tier to its own position right after the two 4-PR contributors, per the count-based ordering established last time. Snapshot bumped to 2026-08-17 / PR #275. CHANGELOG entry added under the same pending version (2.0.11).
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
3 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.
Solved issue - closes #92
Add explainers/precision-recall-curve.md: why ROC/AUC's own
arithmetic stays lenient about rare positives (false positive rate's
denominator is dominated by true negatives) while precision's
denominator - all positive predictions - is much smaller under class
imbalance, letting precision collapse at a threshold AUC never flags.
Checked actual base rates across all seven audits rather than assume
one: Healthcare Readmission's 11.2% positive rate is the most skewed,
so it's the concrete example, anchored to the real frozen baseline
numbers (88.7% accuracy, 0.62 AUC, 0.039 F1 - paper/results-frozen/
results_performance.csv). Detection code (per-group average
precision) tested against a simulated dataset including the
one-class-per-group edge case.
Wired into assets/explainers-data.json, regenerated the .html page,
OG images, sitemap.xml, and llms-full.txt via the real build
pipeline. Updated the hand-maintained spots too: README.md (table,
count 44->45, repo tree), CONTRIBUTING.md's table, ROADMAP.md (Phase
1 checklist + count), METRICS.md's badge, and llms.txt.