Add quantile regression with pinball loss to EBMs#663
Open
Erik-BM wants to merge 2 commits intointerpretml:mainfrom
Open
Add quantile regression with pinball loss to EBMs#663Erik-BM wants to merge 2 commits intointerpretml:mainfrom
Erik-BM wants to merge 2 commits intointerpretml:mainfrom
Conversation
Erik-BM
force-pushed
the
add-quantile-regression-pinball-loss
branch
from
35d9593 to
fdc0486
Compare
tfoslid
approved these changes
Apr 13, 2026
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #663 +/- ##
==========================================
- Coverage 66.40% 66.39% -0.02%
==========================================
Files 75 75
Lines 11589 11589
==========================================
- Hits 7696 7694 -2
- Misses 3893 3895 +2
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
Adds a new "quantile" objective for ExplainableBoostingRegressor that enables quantile regression via pinball loss. Usage: objective="quantile:alpha=0.5" where alpha in (0,1) selects the target quantile. This is useful for prediction intervals, asymmetric risk, and robust median regression. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Signed-off-by: Erik B. Myklebust <erik.bryhn.myklebust@gmail.com>
Erik-BM
force-pushed
the
add-quantile-regression-pinball-loss
branch
from
fdc0486 to
24612e9
Compare
Collaborator
|
Hi @Erik-BM -- The code looks good. It can be made a bit more efficient though because the hessian is always 1.0. To make that, change: OBJECTIVE_BOILERPLATE(PinballRegressionObjective, MINIMIZE_METRIC, Objective_Other, Link_identity, true) to OBJECTIVE_BOILERPLATE(PinballRegressionObjective, MINIMIZE_METRIC, Objective_Other, Link_identity, false) And then you can eliminate the function CalcGradientHessian The notebook is great to have, but could you also add a test. The documentation error is due to a deprecation of the URL that we used to get the adult dataset from -- this has been fixed in the latest commit on main. No need to merge that though since I know why it's happening. Thanks! |
Remove the CalcGradientHessian function from PinballRegressionObjective since the OBJECTIVE_BOILERPLATE now uses k_bHessian=false, making it gradient-only like RMSE. Add tests for quantile regression covering quantile ordering (q10 < q50 < q90), default alpha, and parameter validation (alpha=0, alpha=1, alpha<0 all rejected). Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Signed-off-by: Erik B. Myklebust <erik.bryhn.myklebust@gmail.com>
Erik-BM
force-pushed
the
add-quantile-regression-pinball-loss
branch
from
4a2f016 to
2fe848c
Compare
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.
Summary
"quantile"objective toExplainableBoostingRegressorimplementing pinball loss for quantile regressionobjective="quantile:alpha=0.5"where alpha in (0, 1) selects the target quantile (default 0.5 = median)Changes
shared/libebm/compute/objectives/PinballRegressionObjective.hpp— C++ objective with identity link, constant hessian, SIMD/GPU supportobjective_registrations.hpp— register the new objective_ebm.py— update regressor docstringdocs/interpret/ebm-internals-quantile-regression.ipynb— notebook covering median regression, prediction intervals, and interpretabilitydocs/interpret/ebm-internals.md— link to new notebookTest plan
build.shand verify compilationpytest python/interpret-core/tests/glassbox/ebm/