Add MEADS notebook (IRT + Radon) and fix RegimeSwitchingModel to NUTS#83
Merged
junpenglao merged 1 commit intomainfrom Mar 30, 2026
Merged
Add MEADS notebook (IRT + Radon) and fix RegimeSwitchingModel to NUTS#83junpenglao merged 1 commit intomainfrom
junpenglao merged 1 commit intomainfrom
Conversation
…NUTS - New `book/algorithms/meads.md`: demonstrates MEADS+GHMC adaptation on a 230-dimensional 2PL IRT model and an 89-dimensional Radon hierarchical model. Uses a clean `meads_inference` helper (MEADS-only) with a lightweight init strategy: single NUTS warmup to the typical set, collect a short trajectory, then spread n_chain starting positions using U(-1,1) jitter scaled by the per-coordinate posterior std from the seed positions. Both models achieve Rhat <= 1.05 (IRT) and Rhat <= 1.02 (Radon). - Fix `book/models/RegimeSwitchingModel.md`: the multi-modal regime-switching posterior (label switching, regime collapse) is unsuitable for MEADS; revert to NUTS with `blackjax.window_adaptation`. Anchoring sigma=[log(3), log(10)] to break regime symmetry yields Rhat=1.00 across all chains. - Update `book/_toc.yml` to add `algorithms/meads.md`. Co-Authored-By: Claude Sonnet 4.6 <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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
New
book/algorithms/meads.md: demonstrates MEADS+GHMC adaptation on two benchmark posteriors from the MEADS paper — a 230-dimensional 2PL IRT model (100 students, 30 items) and an 89-dimensional Radon hierarchical model (85 counties). The notebook uses a cleanmeads_inferencehelper (MEADS-only) with a two-step initialization: single NUTS warmup to the typical set, collect a short trajectory, then spread ensemble starting positions usingU(-1,1)jitter scaled by the per-coordinate posterior std. Both models achieve Rhat ≤ 1.05 (IRT) and Rhat ≤ 1.02 (Radon).Fix
book/models/RegimeSwitchingModel.md: the multi-modal regime-switching posterior (label switching, regime collapse) is unsuitable for MEADS — it was reverted to plain NUTS withblackjax.window_adaptation. Anchoringsigma = [log(3), log(10)]at initialization breaks regime symmetry and yields Rhat = 1.00 across all 8 chains.book/_toc.yml: addalgorithms/meads.mdbetweenmclmcandlaps.Test plan
book/algorithms/meads.mdexecutes end-to-end without errorsbook/models/RegimeSwitchingModel.mdexecutes end-to-end with Rhat = 1.00🤖 Generated with Claude Code