feat: per-iteration evidence re-optimization for the interferometer LM engine#628
Merged
Merged
Conversation
…ngine IterFitDpsiSrcInterferometer gains reg_optimize_every: every N accepted LM iterations (and at start) the regularization strength multipliers (a_src, a_dpsi) are re-optimized by maximising the Laplace evidence at the fixed normal equations (one Cholesky per candidate on a log grid) — the objective control of Koopmans 2005 / Vegetti & Koopmans 2009 that stops the source block absorbing potential perturbations. log_evidence made scale-consistent. Campaign local-tier validation: iterative dkappa recovery corr 0.13 -> 0.41, peak 1.20" -> 0.44" (PyAutoLens#627). Co-Authored-By: Claude Fable 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Adds
reg_optimize_everytoal.pc.IterFitDpsiSrcInterferometer(#627, parent #623): every N accepted LM iterations (and at the start) the regularization strength multipliers (a_src, a_dψ) are re-optimized by maximising the Laplace evidence at the fixed normal equations — each candidate costs one Cholesky solve on a log-spaced grid, so the control is nearly free. This is the per-iteration objective control of Koopmans 2005 / Vegetti & Koopmans 2009, and it directly addresses the campaign's characterized failure mode (the source block absorbing the potential perturbation at fixed coefficients).Campaign validation (local tier, table on #627): iterative δκ recovery improves from corr 0.13 / peak 1.20″ (best fixed-coefficient result across a 1000× sweep) to corr 0.41 / peak 0.44″ with
reg_optimize_every=1.log_evidencemade scale-consistent with the adopted multipliers.API Changes
Added only:
reg_optimize_every/reg_optimize_gridparameters (default off — behaviour unchanged) and thereg_scalesattribute.See full details below.
Test Plan
test_autolens/potential_correctionsuite: 51 passed (default path unchanged).test_autolenssuite: 439 passed.--reg-optimize 1(results above).Full API Changes
Added
al.pc.IterFitDpsiSrcInterferometer(..., reg_optimize_every=None, reg_optimize_grid=5)— evidence re-optimization of regularization strengths at fixed F, D;_optimize_regularization(F, D),_effective_regularization(),reg_scales.Migration
Generated by the PyAutoLabs agent workflow.