Repository navigation
examples: params files to current practice; examples/README.md - #508
Merged
Merged
Conversation
Params files are templates: published values, short comments, no narrative. The derivations and history that sat inline move into each example's README (new: hat3/README.md); stale comments that described removed behavior go. ogle0383 loses its init_scale entries, which are not a user key (stripped with a warning). Every edited file parses to the same YAML except that removal, and every edited config builds to a bit-identical start logp (PR body has the table). examples/README.md (new): one entry per example with what it demonstrates, how to run one, and the warning that a fit run in place writes into the directory. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01S5pG2s9aJEo6Xqk6hWvnkW
kelt4_rv+transit+sed: the EXPERT/HIRES/TRES jitter_variance lower bounds sat below the component's validity floor, which replaced them with a warning; the floor is the bound either way. ob08092: `sigma: 0` on star.Source.mass, a derived parameter there, had no effect (warned). Start logp, free RVs and user-start misses are bit-identical; the warnings go. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01S5pG2s9aJEo6Xqk6hWvnkW
tests/test_sed_grid_barrier.py pins it there: the caveat belongs where the person running the example sees it. Comment-only; the YAML is unchanged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01S5pG2s9aJEo6Xqk6hWvnkW
The entry widened the SED's loggsed barrier while GJ 1214 (logg 5.022) sat past the NextGen grid's old logg 5.0 ceiling. The tables reach 5.5 since 2026-10-20, so the star is inside the grid and the measured barrier applies. Start logp +2.2e-4 (108507.98057121364 -> 108507.98079425778), free RVs and user-start misses unchanged; approved by JDE 2026-10-28. The README keeps the history; sed.md, run.md and mkparam.py describe the entry in the past tense. test_gj1214_example_ships_the_softened_barrier pinned the entry and goes with it; the barrier-softening mechanics stay covered by the synthetic tests in the same file and by test_mkparam's bound_scale round trip. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01S5pG2s9aJEo6Xqk6hWvnkW
hat3.params.yaml and hat3_transitonly.params.yaml carried star mass 0.92 +/- 0.05 beside the Torres relations, under a comment saying it was only there for want of a mass constraint and should come out of a final run. Approved by JDE 2026-10-28. Start logp: hat3 272773.2040354589 -> 272773.23110895697 (+0.0271, the Gaussian term at the start mass 0.9085); hat3_transitonly unchanged (its start sat on the prior center). Free RVs and user-start misses unchanged. hat3_staronly keeps its prior (no mass relation there). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01S5pG2s9aJEo6Xqk6hWvnkW
`cd examples/kelt4rvonly && poetry run exozippy kelt4.yaml` pointed at a directory that does not exist; the RV-only example is examples/kelt4/kelt4_rvonly.yaml. Verified as written (in a scratch copy of examples/): the fit runs to completion, rc 0. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01S5pG2s9aJEo6Xqk6hWvnkW
The sweep2/, sweep3/, sweep4/, ab047/loggwide and ab194/{mamajek,mistlens}
configs named an events/<NNN>/DC2018_<NNN>_seed.json under the
`mmexofast:` key, which was removed on 2026-10-01 and raises at load.
Approved by JDE 2026-10-28. Only that line is deleted: each config now
builds (prepare + build_model) with the built-in peak finder seeding
t_0/u_0/t_E, 0 user-start misses; the MMEXOFAST solutions those JSONs held
are in the events/<NNN>/DC2018_<NNN>_seed.params.yaml fragments and are
not merged here. No generator writes the key. Start logp per config is
in the PR body (there is no "before": they raised).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S5pG2s9aJEo6Xqk6hWvnkW
jdeast
force-pushed
the
docs/examples-readme-sweep
branch
from
October 28, 2026 21:30
c09ea01 to
10f2fe7
Compare
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
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.
What this does
A. Example params files swept to current practice. A params file is a template: published values, short comments, no narrative. Every edit here is either prose that moved into the example's README, a stale comment removed, or an entry the build already overrides or ignores (with a warning). No value, seed, per-seed initval, citation or modeling choice changed, except the approved changes listed under "Additional approved changes" (each its own commit).
_linear), ob170114, ob161045, ob140939 (+_today), ob161003, KMT-2021-BLG-1122L, kb180087_obj3, KMT-2019-BLG-1806, DC2018_128hat3,_mist,_mist_claret,_transitonly)hat3/README.md(also a table of the five configs). Own commit (approved by JDE): the hand mass prior (0.92 +/- 0.05) removed fromhat3andhat3_transitonly, where the Torres relations constrain the mass;hat3_staronlykeeps it.star.A.loggsed: {bound_scale: 25.0}REMOVED (own commit; approved by JDE, start logp +2.2e-4): it softened the SED barrier while the star sat past NextGen's old logg 5.0 ceiling, and the tables reach 5.5 now. The README keeps the history; sed.md, run.md and mkparam.py describe it in the past tense;test_gj1214_example_ships_the_softened_barrier, which pinned the entry, is removed (the softening mechanics stay covered by the synthetic tests in that file and test_mkparam's round trip).init_scaleremoved from 7 entries (not a user key: stripped with an "obsolete and ignored" warning); header comment fixed.sigma: 0dropped onstar.Source.mass(derived here: "sigma=0 has no effect", warned).jitter_variancelower bounds removed: each sat below the component's validity floor, which replaced it with a warning. Stale mass-prior history trimmed.gaussian_width,upper_bound, the undefined EXPERT/FIES entries) removed; the orbit.BC.period history trimmed.Every edited params file parses to the identical YAML except the intended removals above (checked with
yaml.safe_loadon both versions).B.
examples/README.md(new). One entry per example (collections as one entry), grouped exoplanet / microlensing / other, with what each demonstrates, how to run one, and the warning that a run in place writes into the directory (fitresults/, andmkparam'sparams.2.yaml).Evidence: start logp per seed, before -> after
Each config built from a scratch copy of the ORIGINAL and the EDITED tree (prepare + build_model, start logp of every seed from
System.get_raw_starts,ModelAuditor.check_user_startsmisses, every exozippy warning), serially, underulimit -v 12000000.Notes: "identical" = every seed's logp equal to the bit, free RVs and misses unchanged. "(unchanged)" = the config's params file was not edited (baseline only). Configs that share a params file with a built one and differ only in sampler settings (
ob09020_{diag,polish,tune,tuneonly},KMT-2019-BLG-1806_nt8long) were not built separately. The kelt4_rv+transit+sed and ob08092 removals were built again on their own: logp bit-identical (84411.3135835002, 6252.18472912271), warnings 5 -> 2 and 1 -> 0. ogle0383's warning count goes 1 -> 0 (the init_scale warning).theophyllinedoes not build in a scratch copy (its data file is fetched, not committed); its params file has no entries before or after.DC2018/ab047/loggwideraised on itsmmexofast:config key before the commit below removed it.Additional approved changes (JDE 2026-10-28), each its own commit
mmexofast:key removed from 21 configs (sweep2/,sweep3/,sweep4/x {047, 069, 103, 152, 194, 223},ab047/loggwide,ab194/{mamajek,mistlens}). They raised at load before, so there is no "before". Only that line is deleted; every one now builds (prepare + build_model), with the built-in peak finder seeding t_0/u_0/t_E and 0 user-start misses. The MMEXOFAST solutions those JSONs held live inevents/<NNN>/DC2018_<NNN>_seed.params.yamland are NOT merged in (no seeds invented). No generator writes the key.lens.Companion.alpha: 1.832is degrees and is left as is. The value came from MMEXOFAST, whose alpha convention is EXOZIPPy's (docs/alpha_conventions.md, item 0; conventions.md C18), and MMEXOFAST writes degrees (DC2018_128's mmexofast.json: -52.15). Measured: the shipped 1.832 deg gives start logp 26739.644; reading it as radians (104.983 deg) gives 25985.957 (-753.7), its +180 twin 25956.931, its negation 25933.696. The data prefer the value as written by >750 nats, and the earlier fit's restart file (OGLE_0383.paramsLD.2.yaml) sits at 1.877 deg.cd examples/kelt4rvonly && poetry run exozippy kelt4.yamlnamed a nonexistent directory; it is nowcd examples/kelt4 && poetry run exozippy kelt4_rvonly.yaml. Verified as written in a scratch copy of examples/: the fit runs to completion (rc 0, results andkelt4.params.2.yamlwritten).The branch is rebased onto master d216e0e (#507, the blackjax fix).
Held (not in this PR)
tests/test_dc18_arch_config.pyon the synthetic event those tests generate: with the orbit block present, the peak finder'sprobe_startreportssource.0.t_0as INFORMED (rank 40 "solved", at the t_0 window's upper edge, 2458509.6) where without it the out-of-window defaults.yaml backstop (2460000, rank 20) is reported uninformed, so the finder "holds" a t_0 nobody gave and the build then refuses the backstop start. Bisected: the orbit block alone causes it (dropping the hand links alone does not). That is an engine/peak-finder interaction, not a generator issue, so the commit is held until it is fixed.Suggestions (modeling choices, not applied)
hat3.params.yaml:band.TESS.u1/u2carryinitval+sigmaon derived coefficients; the start never reaches the model (2 user-start misses). Amu:would state the prior center without implying a start.kelt17_full.params.yaml:planet.b.pandplanet.b.radiusdisagree slightly (p dropped, rel 3e-4);star.A.vmacro/vbeta/vmicro(RM) andstar.A.vline(DT) are unused by the DT-only / RM-only configs that share the file (warned).KMT-2021-BLG-1122L: the LensB/LensC hand links on distance/pm/rv could become atype: staticorbit (pending the issue above); LensC could now be seeded withs:/alpha:(per-companion relations exist since 8.6.13) -- the README still says slot 1 ignores them; the zeropointmu16/19 workaround for the diazp_refmay be revisitable.ob140939_today: the av/feh/logmass priors could carrycitation:(Surot+2020, Ness+2013/Zoccali+2017).star.Lens.radius/teff/feh, which those configs deactivate ("your sigma is DROPPED").ob09020: six unfiltered (U) light curves still warn that the flux bootstrap finds a negative blend although the params file seeds their fluxes.🤖 Generated with Claude Code
https://claude.ai/code/session_01S5pG2s9aJEo6Xqk6hWvnkW