Skip to content

fix: un-park pixelization/MGE scripts (stale NEEDS_FIX, now green)#141

Merged
Jammy2211 merged 2 commits into
mainfrom
feature/pix-inversion-not-positive-definite
Jul 21, 2026
Merged

fix: un-park pixelization/MGE scripts (stale NEEDS_FIX, now green)#141
Jammy2211 merged 2 commits into
mainfrom
feature/pix-inversion-not-positive-definite

Conversation

@Jammy2211

Copy link
Copy Markdown
Collaborator

Summary

Removes 3 stale NEEDS_FIX markers from config/build/no_run.yaml. The
LinAlgError: matrix not positive definite (pixelization) and matrix singular
(MGE) failures were fixed the same day the markers were filed by the
PyAutoArray GaussianKernel PD-guarantee (f1817af0 — trace-scaled diagonal
jitter + symmetrisation on the regularization matrix); the markers were simply
never removed. No source or script changes are needed — this is marker cleanup
with proof the scripts are green.

Verification (reproduce-or-refute):

  • All 3 scripts run GREEN end-to-end in the exact build-sweep mode that
    births the markers (PYAUTO_TEST_MODE=2, real data, PYAUTO_SMALL_DATASETS
    unset). The MGE np.linalg.solve is not behind the sampler, so it is
    genuinely exercised.
  • 40-draw numpy inversion A/B across the entire GaussianKernel LogUniform
    prior (coefficient/scale 1e-6 -> ~5e5): 0 raises, 0 non-finite on both
    cholesky (default) and slogdet.

Addresses Heart RED reason "58 stale parked script(s)" (corrective PR).

Scripts Changed

No script edits -- config/build/no_run.yaml only. Markers removed (scripts now un-parked and verified green):

  • imaging/features/pixelization/modeling -- was LinAlgError: matrix not positive definite
  • interferometer/features/pixelization/modeling -- was LinAlgError: matrix not positive definite
  • interferometer/features/multi_gaussian_expansion/likelihood_function -- was LinAlgError: matrix singular -> InversionException

Unrelated NEEDS_FIX/SLOW markers (imaging/modeling KeyError, shapelets SLOW) left intact.

Test Plan

  • Each un-parked script verified green end-to-end (PYAUTO_TEST_MODE=2, real data)
  • 40-draw numpy inversion A/B across full prior -- no raises / non-finite

Refs #140 (workspace half; HowToGalaxy PR paired).

Generated by the PyAutoLabs agent workflow.

Jammy2211 and others added 2 commits July 21, 2026 13:40
The imaging/data_preparation/manual/mask_irregular script now runs green on
clean main (PASSED 8.6s via the build's own execute_script runner). The
2026-04-10 'silent failure' was Convolver.from_gaussian API drift, since fixed
by the PyAutoArray convolver consolidation (#360/#361). Folded into this task
per the mask_irregular_silent_failure prompt (all 4 imaging repos, one root
cause). autogalaxy_workspace + HowToGalaxy handled here; autolens_workspace +
HowToLens ride slam-adapt-inversion-cascade (#300).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The LinAlgError 'matrix not positive definite' (pixelization) and 'matrix
singular' (MGE) markers were fixed the same day they were filed by the
PyAutoArray GaussianKernel PD-guarantee (f1817af0) and never removed. Verified
all three scripts run green end-to-end in the build-sweep mode (PYAUTO_TEST_MODE=2,
real data), plus a 40-draw numpy inversion A/B across the full LogUniform prior
(0 raises, 0 non-finite).

Removes 3 stale NEEDS_FIX markers; unrelated markers left intact.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
@Jammy2211 Jammy2211 added the pending-release PR queued for the next release build label Jul 21, 2026
@Jammy2211

Copy link
Copy Markdown
Collaborator Author

Folded in (concurrent session): a chore: un-park mask_irregular no_run marker commit rides along on this branch. imaging/data_preparation/manual/mask_irregular runs green on main (PASSED 8.6s) — its 2026-04-10 'silent failure' was Convolver.from_gaussian drift, fixed by PyAutoArray #360/#361. Same marker-only, config-only class as this PR. Combined diff touches config/build/no_run.yaml only.

@Jammy2211
Jammy2211 merged commit 3c0c7e4 into main Jul 21, 2026
8 checks passed
@Jammy2211
Jammy2211 deleted the feature/pix-inversion-not-positive-definite branch July 21, 2026 13:07
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

pending-release PR queued for the next release build

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant