Add order statistic diagnostic for compare()
#237
Conversation
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
jordandeklerk
commented
Oct 30, 2025
jordandeklerk
commented
Oct 30, 2025
Comment on lines
+584
to
+589
| if candidate_sd == 0 or not np.isfinite(candidate_sd): | ||
| warnings.warn( | ||
| "All models have nearly identical performance.", | ||
| UserWarning, | ||
| ) | ||
| return |
There was a problem hiding this comment.
Not sure if we want to add more to this warning? This is mostly to avoid the Scipy runtime error that will happen if sd=0. Probably not very realistic in practice, but theoretically possible.
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #237 +/- ##
==========================================
+ Coverage 84.72% 84.78% +0.05%
==========================================
Files 41 41
Lines 4950 4974 +24
==========================================
+ Hits 4194 4217 +23
- Misses 756 757 +1 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
Documentation build overview
Show files changed (12 files in total): 📝 12 modified | ➕ 0 added | ➖ 0 deleted
|
aloctavodia
reviewed
Oct 31, 2025
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Outdated suggestions cannot be applied.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.
This implements the order statistic diagnostic for detecting selection induced bias when comparing many models in
compare(). When more than 11 models are compared using full PSIS-LOO-CV, the diagnostic estimates whether the observed performance difference could be due to chance by comparing the best model's ELPD difference against the expected maximum order statistic under a null hypothesis of equal performance.This diagnostic is not done for the subsampling case. As far as I can tell, it isn't done here either https://github.com/stan-dev/loo/blob/d6fe380161fcd3ba07065ce0a525146abdb2c1d7/R/loo_compare.psis_loo_ss_list.R. I'm not exactly sure why this is, but my guess is that the the theoretical assumptions underlying the test don't account for the additional variance and approximation bias introduced by subsampling, making it unclear whether the null distribution would be properly calibrated in that setting.
Resolves #234