mcsamplerGPU: prevent adaptive-proposal support truncation - #39
Merged
oshaughnessy-junior merged 1 commit intoAug 1, 2026
Merged
Conversation
On a mild 2D Gaussian test (T4b, floor_level unset), the adapted sampler was
biased by -0.32 nats while every within-run error estimate read ~0.02: the
1D marginals of the drawn samples matched the claimed p_s exactly, but
E[prior/p_s] = 0.37, i.e. ~63% of the prior volume sat in histogram bins with
exactly zero proposal probability. A zero bin is an absorbing state (it can
never be re-drawn), so the sampled support shrinks irreversibly and the
integral silently loses the mass outside it.
Four fixes:
- compute_hist: clamp the uniform-mixture floor to HIST_FLOOR_LEVEL_MIN=1e-2
so no bin can reach zero probability (production ILE already passes 0.1;
the clamp only binds for smaller/unset floors).
- integrate_log: build adaptation weights from the stored tempered importance
weights exp(tempering_exp*lnL + ln p - ln p_s) so the weighted histogram
estimates the fixed target L^beta * prior -- the documented contract
(see integrate() and the ILE driver comment). The old lnL+max(maxlnL,200)
weights ignored tempering_exp and 1/p_s; being near-flat, each histogram
replayed the previous proposal's sampling noise, a multiplicative random
walk that collapsed the proposal onto a comb of surviving bins (61/100 per
dim in the test) and drove the truncation above.
- integrate_log: n_adapt freeze test double-multiplied by n (n_adapt was
already scaled at parse time), so adaptation never froze regardless of the
requested chunk count; also scale the no-kwarg default consistently.
- pdf_from_hist: clamp bin index to n_bins-1 (right-edge sample previously
indexed out of range).
Validation (2D Gaussian, n=2000, neff=1000, 16 runs/config): bias
-0.321 +/- 0.011 -> -0.003 +/- 0.004 for all n_adapt in {5..100} and
tempering_exp in {0, 0.1, 1.0}; final proposal support 56% -> 100%;
runs now reach neff~1000 instead of exhausting nmax at neff~350.
E[prior/p_s] over fresh draws from the adapted proposal: 0.373 -> 1.0002.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
oshaughnessy-junior
deleted the
codex/mcsampler-gpu-adaptive-support-fix
branch
August 13, 2026 14:42
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.
Context
PR #38 superseded #37 and landed the MC-error estimator, but the separate validated GPU adaptive-proposal fix ff0a04b was not included. This PR extracts only that commit onto current rift_O4d.
Changes
The original validation found a 2D Gaussian bias of -0.321 +/- 0.011 nats before the fix and -0.003 +/- 0.004 afterward, with proposal support restored from 56% to 100%.
Validation
Both focused sampler runs pass on CPU. This host has no CuPy device; the original commit records the GPU/device-level validation.