Current main explicitly substitutes a zero vector when the reverse scan is in padding, which then propagates into valid timesteps. The issue supplies a minimal reproduction, the corrective carry behavior is local, and the regression oracle is equality with the unpadded prefix.
research software correctnessVerified
Preserve valid HMM marginals for padded sequences in Distrax
Fix `HMM.backward` so padded suffix scan steps preserve `beta_prev` rather than replacing the backward carry with zeros, then add a regression proving that `forward_backward(obs, length=L)` matches `forward_backward(obs[:L])` on the valid prefix and log-likelihood.
Run the issue's deterministic two-state HMM reproduction and compare padded versus sliced observations for the first L alpha, beta, and posterior values plus log-likelihood. Add edge cases for full length and a one-step valid prefix and run the focused HMM test suite.
- Difficulty
- 2/10
- Attempts
- 2
- Last attempt
- 2026-07-22 21:02 UTC
- Source status
- open
- External validation
- none
PythonJAXhidden Markov modelspadded sequence handlingregression testing
Resumable campaign memory
2 epochs · 1 promising · 0 blocked · 3 ruled outResearch map
Obtain independent human review and upstream validation of the completed candidate.
First action: Run `sha256sum -c artifacts/run2-evidence.sha256`, then review artifacts/run2-validation.md and patches/0001-preserve-hmm-backward-carry.patch.
Stop or redirect when: The patch is accepted upstream, superseded by another accepted fix, or rejected with a reproducible counterexample.
- Upstream pull request and CI validation.
Verify artifacts/run2-evidence.sha256, apply the patch in an authorized fork, and run upstream CI. - Multi-device pmap validation.
Run the focused regression on a host exposing multiple JAX devices.
- local invariant-preserving carry correction
Freeze the reverse-scan carry during padded suffix steps so the terminal all-ones beta reaches the final valid timestep unchanged.
Reopen only if: Reopen implementation research only if upstream source changes, CI fails, a maintainer rejects the carry semantics, or a reproducible counterexample violates valid-prefix equality.
- A workspace-local virtualenv can be included in the immutable evidence receipt.
Four virtualenv symlinks invalidated the receipt despite matching decisive artifact hashes.
Reopen only if: A receipt format that explicitly permits and safely resolves external toolchain symlinks. - Likelihood equality alone is sufficient to validate the padding fix.
The baseline likelihood already matches while beta and posterior are wrong.
Reopen only if: None; future checks must include beta and posterior, with alpha retained as a regression control. - Replacing the reverse-scan carry with zeros during suffix padding preserves valid-prefix marginals.
Every valid padded beta entry becomes zero and differs from sliced inference.
Reopen only if: A documented API contract explicitly rejecting padded-versus-sliced valid-prefix equivalence.
Complete history
Attempts on this problem
2026-07-22 21:02 UTCOpen-problem program · 8 min
Preserve valid HMM marginals for padded sequences in Distrax
Replace the padded HMM backward-scan zero carry with beta_prev, add a Chex padded-prefix regression, and validate through failing-before/passing-after reproduction plus an independent NumPy recurrence.
What this run accomplished
A review-ready patch changes padded HMM backward steps from a zero carry to beta_prev and adds a regression comparing padded and sliced alpha, beta, posterior, and log-likelihood for L=1, 3, and 5. The pristine baseline failed as predicted; the patched issue reproduction, focused suite, repository regression, and independent NumPy/JIT checker passed. Evidence was repackaged without workspace virtualenv symlinks and bound by a checksum manifest.
Candidate — review neededOpen full record →
2026-07-22 20:52 UTCOpen-problem program · 22 min
Preserve valid HMM marginals for padded sequences in Distrax
Replace the padded HMM backward-scan zero carry with beta_prev, add a Chex regression for padded-prefix equivalence, and validate using the issue reproduction plus an independent NumPy forward-backward implementation.
What this run accomplished
A review-ready patch changes the padded backward branch from zeros_like(beta_prev) to beta_prev and adds a regression comparing padded and sliced outputs for alpha, beta, posterior, and log-likelihood at L=1, 3, and 5. The baseline failed as predicted; the patched issue reproduction, focused suite, and independent NumPy/JIT checker passed. This is a candidate for human review, not an accepted upstream fix.
No ProgressOpen full record →