feat(metrics): add ECE and RMS calibration error metrics#3913
Open
shadowmodder wants to merge 1 commit into
Open
feat(metrics): add ECE and RMS calibration error metrics#3913shadowmodder wants to merge 1 commit into
shadowmodder wants to merge 1 commit into
Conversation
Closes EleutherAI#2189 Adds two new calibration quality metrics: - ece: Expected Calibration Error with equal-width binning (Guo et al. 2017) - rms_ce: Root-Mean-Square Calibration Error Both are registered as output_type=['multiple_choice'] metrics compatible with the brier_score pattern. Each accepts (gold_label, predicted_probability) pairs and returns a scalar in [0, 1]. Lower is better.
|
|
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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 #2189
Summary
Adds two new calibration quality metrics as requested in #2189:
ece— Expected Calibration Error with equal-width binningrms_ce— Root-Mean-Square Calibration ErrorBoth follow the existing
brier_scorepattern exactly:@register_aggregation+@register_metricoutput_type=["multiple_choice"](gold_label, predicted_probability)pairs[0, 1]. Lower is better.Design
ECE bins predictions into
n_binsequal-width buckets and computes the weighted average of |accuracy − confidence| per bin (Guo et al., "On Calibration of Modern Neural Networks", ICML 2017).RMS-CE uses an L2 norm instead of L1, giving more weight to large miscalibration gaps.
Usage in a task YAML
References