maintenance: validate blackjax 1.6.2 + mainline NSS smoke + NUTS API check - #143
Merged
Conversation
…142) Phase 0(b) / the CPU environment half of CP-2, per human direction (install/API validation only; GPU MGE smoke and local/RAL venv upgrades remain laptop work): - blackjax 1.6.2 + jax 0.10.2 (CPU) + released autofit 2026.7.29.1 coexist cleanly; PyAutoFit pins blackjax>=1.2.0 (floor, no cap, in the 'optional' extra) so the Gate-A floor bump has no cap to fight — but blackjax is not a hard autofit dependency, so upgrades must install it explicitly. - scripts/misc/searches/nss_smoke.py: rerunnable CPU smoke for every environment's upgrade. blackjax.nss 2D toy exercises the full mainline API (native-space logprior_fn, num_delete, dlogz termination on integrator.logZ_live - logZ, ns.utils.finalise + log_weights ensemble): logZ -4.609 +/- 0.069 vs analytic -4.605, posterior recovered. af.BlackJAXNUTS runs on 1.6.2 unmodified (0 divergences, acc 0.92). - results/notes/inference/phase_00_unblocking/RESULTS.md: the findings, incl. verified-in-source provenance for the H2.1 inner-steps/logZ-bias guidance; PROGRAMME phase table 0(b) -> validated (cloud CPU). Co-Authored-By: Claude <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018dV9h5h76hZ6trGZKWJFHe
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
2 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Closes #142. Phase 0(b) of the inference programme — the CPU environment half of critical-path item CP-2, run in a cloud session per human direction ("just getting install type stuff working"). The GPU MGE smoke half of CP-2 and the actual local-venv/RAL upgrades remain laptop work.
Results (all PASS)
pip checkclean). Two dependency facts recorded: PyAutoFit pinsblackjax>=1.2.0— floor only, no cap, and only in theoptionalextra — so the Gate-A floor bump is a one-liner; and blackjax is not a hard autofit dependency, so upgrades mustpip install -U blackjaxexplicitly.blackjax.nss2D toy smoke: exercises the exact mainline API surface Phase 2's profiling-local runner will wire (native-spacelogprior_fn,num_delete, dlogz termination viastate.integrator.logZ_live − logZ,ns.utils.finalise+log_weightssimulated-volume ensemble,ns.utils.sample). logZ = −4.609 ± 0.069 vs analytic −4.605; posterior mean/σ recover truth. Bonus provenance: the installed 1.6.2 docstring states thenum_inner_steps ≥ max(5, 2·d)rule and the upward logZ-bias direction verbatim — H2.1's citation is now verified-in-source.af.BlackJAXNUTSon 1.6.2: runs unmodified (window_adaptation,nuts,diagnostics.effective_sample_sizeall intact) — toy posterior recovered, 0 divergences, acceptance 0.92, ESS_min 244. No PyAutoFit compatibility fix needed; the §7 Phase-6 items stay features, not fixes.Contents
scripts/misc/searches/nss_smoke.py— rerunnable smoke (exits non-zero on FAIL, guards on blackjax < 1.6); run it in each environment after its upgrade.results/notes/inference/phase_00_unblocking/RESULTS.md— the findings record.ruff check/ruff format --check/build_readme.py --checkall pass. CPU toy timings carry no profiling meaning (wrong tier, per the programme's own rules).🤖 Generated with Claude Code
https://claude.ai/code/session_018dV9h5h76hZ6trGZKWJFHe
Generated by Claude Code