ILE: derive the psi and phi_orb priors from their sampling ranges (evidence change: -ln 2) - #190
Open
oshaughnessy-junior wants to merge 1 commit into
Conversation
The ILE drivers sample psi on (0, 2 pi), so that --internal-rotate-phase can use (phi+psi, phi-psi) coordinates, but passed the fixed constant mcsampler.uniform_samp_psi = 1/pi, written for [0, pi), as the psi prior. A prior defined independently of the range it is integrated over has no reason to be normalized: this one had mass 2, and mass 4 under --internal-rotate-phase (range (0, 4 pi)), where phi_orb's fixed 1/(2 pi) prior doubled as well. Every reported lnZ carried +ln 2 (+ln 8 rotated). Fix: capture psi_prior_range and phi_orb_prior_range once the sampling ranges are final and build both priors with ret_uniform_samp_vector_alt over those ranges, in the batchmode and legacy drivers. The likelihood is periodic in both angles and each range covers whole periods, so lnZ is now that of a prior normalized on one period. EVIDENCE CHANGE. Every ILE lnZ drops by ln 2 (ln 8 with --internal-rotate-phase). Posteriors are unchanged. Old all.net or composite tables cannot be mixed with new ones without adding the offset. Backport of the rift_O4d change; test_angle_prior_normalization.py carries the constructor and driver-wiring checks. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
oshaughnessy-junior
deployed
to
private-review-dispatch-rift-upstream
September 9, 2026 10:12 — with 
GitHub Actions
Active
GitHub Actions
Active
mebiri
pushed a commit
to mebiri/research-projects-RIT
that referenced
this pull request
Sep 15, 2026
…nterp-cubic-and-grid-fix Two shipped scripts that cannot do what they document (--interp cubic unreachable; util_ManualOverlapGrid --inj broken)
mebiri
pushed a commit
to mebiri/research-projects-RIT
that referenced
this pull request
Sep 15, 2026
Only conflict: .travis/test-jax.sh EXPECTED_TESTS. Neither side's number was right for the MERGED manifest -- 61 counted this branch's file list, 30 counted rift_O4d's, which lacks test_jax_fairdraw_export.py but adds test_interp_choices.py from oshaughn#190. Resolved to 64 by RUNNING collection on the merged FILES list, not by arithmetic on the two sides. rift_O4d added test_interp_choices.py to FILES without an entry in the per-file count ledger above EXPECTED_TESTS; added (3), so that block remains a complete accounting rather than a partial one that happens to sum short. Checked that the auto-merge of bin/integrate_likelihood_extrinsic_jax kept both sides rather than trusting it: this branch's 2-arg fairdraw_size, both length-mismatch guards and the corrected --n-fairdraw-extrinsic-samples help text are present, and oshaughn#190's `--interp cubic` choice is present. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
mebiri
pushed a commit
to mebiri/research-projects-RIT
that referenced
this pull request
Sep 15, 2026
…interp cubic) oshaughn#180 has landed, so this PR now targets rift_O4d directly rather than stacking. Three conflicts: * --interp cubic. oshaughn#190 shipped the same fix independently, with a help string explaining that linear undershoots the rholm peak and biases the recovered arrival time. Took THEIRS wholesale -- it is a strict superset of the bare `choices` widening this branch carried -- and dropped mine. * The gate's per-file ledger. Both entries are needed (this branch's chooser file, oshaughn#190's test_interp_choices.py); kept both. * EXPECTED_TESTS. Resolved to 95 by RUNNING collection on the merged FILES list. 64 (rift_O4d) + 31 (chooser) = 95 confirms nothing else shifted. TWO STALE CLAIMS OF MY OWN, found by reading what landed rather than only the conflict hunks: * The gate ledger entry for test_jax_tempering_chooser.py still said the chooser is "actually assigned to opts.adapt_weight_exponent" -- the contract REMOVED in cf02b9d when that write turned out to make event 1 of a batch read event 0's choice. Corrected to the return-value contract. * The --smc-puffball branch comment said smc_puffball_sample "returns post_weight uniform with temper=1.0". oshaughn#180 changed that: it now reports the inv_T its ladder actually reached and a matching L^(1-inv_T) weight, and its export weights are uniform only because analyze_one REFUSES the event unless the ladder finishes at inv_T=1. Skipping the guard is still correct -- the signature still swallows `temper` in **_ignore, so the exponent is genuinely inert -- but the comment now says why, and warns against re-deriving "its weights are always uniform" from that branch. Checked the driver's automerge kept both sides rather than trusting it: oshaughn#180's 2-arg fairdraw_size and both length-mismatch guards are present alongside this branch's resolve_tempering_exponent, the resolved_beta wiring at both sampler call sites, and a single _USABLE_EXPORT_ESS shared with fairdraw_indices. 65 tests pass across the two interacting files (31 chooser + 34 export). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
mebiri
pushed a commit
to mebiri/research-projects-RIT
that referenced
this pull request
Sep 15, 2026
Caught reading PR oshaughn#186's own diff after the rift_O4d merge. I wrote "79" when 79 was current; merging oshaughn#180 (64) plus this branch's chooser took the floor to 95 and left my own note stale -- the same failure the note itself documents, one revision later. Also records where the count came from (27 -> 48 -> 64 via oshaughn#180 -> 95 here, with oshaughn#190's test_interp_choices.py along the way) and re-measures the wall time on the host it was actually run on (859 s, ldas-pcdev11, 16 cores), so the 60-minute timeout is justified against a current number rather than a stale one. Co-Authored-By: Claude Opus 5 <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.
Evidence change: every ILE lnZ drops by ln 2
The ILE drivers sample psi on (0, 2 pi) so that
--internal-rotate-phasecan use (phi+psi, phi-psi) coordinates. The psi prior was the fixed constantuniform_samp_psi = 1/pi, written for [0, pi) and defined independently of that range, so it integrated to 2. Under--internal-rotate-phasethe range is (0, 4 pi) and the fixed phi_orb prior doubled too: mass 8. Measured with--zero-likelihoodon this driver, lnZ = 0.702 +/- 0.009 (ln 2 = 0.693) and 2.094 +/- 0.009 (ln 8 = 2.079); after the change 0.003 and 0.008 (time-marginalized run on a synthetic zero-signal frame, default sampler).This PR derives both angle priors from the sampling ranges the driver actually uses, in the batchmode and legacy drivers. The likelihood is periodic in both angles and every range covers whole periods, so lnZ is now that of a normalized prior. Posteriors are unchanged.
Breaking for table reuse. Old
all.netor composite tables cannot be mixed with new ILE rows: the old rows sit ln 2 high (ln 8 under rotate-phase). Reused CIP fits inherit the same constant.Backport of the same change on
rift_O4d(oshaughnessy-junior#306), where the JAX driver already used a normalized psi prior and the two drivers now agree. Tests:test_angle_prior_normalization.py(constructor and driver wiring), 5/5 locally.🤖 Generated with Claude Code