feat(forecaster): quantile-LightGBM probabilistic head factory (#3)#349
Merged
Conversation
Implements chapter-16 "adopt next" #3 (point → calibrated bands) as a thin, sf2-safe convenience layer over the EXISTING interval machinery — it does not duplicate predict_interval / predict_quantiles (residual bootstrapping + split-conformal already on ForecasterRecursive). Adds native quantile regression heads instead. - quantile_lgbm_forecaster_factory(config, quantiles=(0.1,0.5,0.9)) → one ForecasterRecursive per quantile with LGBMRegressor(objective="quantile", alpha=q), same lag/rolling config as the default factory. Deterministic, LightGBM-only. Refs hong16b, roma19a. - predict_quantile_band(forecasters, steps, enforce_monotonic=True) → assembles the per-quantile forecasts into q_<level> columns and applies the Chernozhukov rearrangement (row-wise sort) to remove quantile crossing. - Fail-safe quantile validation (open (0,1), unique, non-empty). For distribution-free coverage, compose with the existing ForecasterRecursive.predict_interval(method="conformal"). Tests: 14 new (factory params, validation, rearrangement, fit/predict integration). Suite 2413 passed; factories.py 100% coverage. Docs regenerated. Co-Authored-By: Claude Opus 4.8 (1M context) <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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Implements chapter-16 "adopt next" #3 — a probabilistic head turning the point forecast into calibrated bands — as a thin, sf2-safe convenience layer.
Non-duplication
The recursive forecaster already has the interval machinery:
predict_interval(residual bootstrapping and split-conformal),predict_quantiles,predict_dist, binned residuals, and CRPS/pinball/coverage metrics. This PR does not reimplement any of that. It adds the one missing piece — native quantile-regression heads — plus a band assembler.What's new (
multitask/factories.py)quantile_lgbm_forecaster_factory(config, quantiles=(0.1, 0.5, 0.9))→ oneForecasterRecursiveper quantile usingLGBMRegressor(objective="quantile", alpha=q), same lag/rolling config as the default factory. Deterministic, LightGBM-only. Refshong16b,roma19a.predict_quantile_band(forecasters, steps, enforce_monotonic=True)→ assembles per-quantile forecasts intoq_<level>columns and applies the Chernozhukov rearrangement (row-wise sort) to remove quantile crossing.(0,1), unique, non-empty).For distribution-free coverage, compose with the existing
ForecasterRecursive.predict_interval(method="conformal").Tests & QC
14 new tests (factory params/validation, rearrangement, fit→predict integration). Suite 2413 passed,
factories.py100% coverage. ruff/black/isort clean, REUSE compliant, docs regenerated (2 new pages).🤖 Generated with Claude Code