perf(ep): EP factor searches skip per-search visuals by default (#1642, PR 2 of 3)

[Jammy2211]

Summary

PR 2 of 3 under #1642, stacked on #1643 (base = feature/ep-factor-search-overhead). Retarget to main after #1643 merges. PR 3 (mapper fast path) follows.

EP re-enters the same factor search once per EP step. At small per-factor fits the per-search plotting dominated the wall time, and the images were overwritten on every step anyway. This PR switches it off by default:

• New config key general.yaml -> output -> visualize_ep_factor_searches: false. It is read with .get(..., False), so user configs that predate the key keep working.
• NonLinearSearch._visualize_fit / _visualize_before_fit are class attributes (default True). optimise (the EP entry point) sets them around self.fit(...) and resets both to True in a finally.
• pre_fit_output gates analysis.visualize_before_fit on _visualize_before_fit, so only the first search of each factor (number == 0) draws before-fit visuals.
• SearchUpdater(visualization_enabled=True): when False, visualize returns immediately (no analysis.visualize / visualize_combined / plot_results, no samples loaded for plotting). The cached _updater re-applies self._visualize_fit on every access.

Samples, summaries and the EP optimiser's own graph.info / graph.png / graph_factors.png / mean_field_evolution.png / ep_history.csv are unchanged.

Harness: n5_profile, repeat 0, max_steps 3, nlive 100, seeded. The machine was shared with another profiling job (load average ~6), so the PR 1 head was re-run interleaved as a load-matched control.

	PR 1 (1678905)	PR 2 (1cd4a5b)
run_nested calls (15 factor searches)	15	15
per-search wall, quiet machine	1.575 s	not measured (target ≤ 1.5 s)
per-search wall, loaded, interleaved pairs	3.52 / 4.12 / 2.01 s	2.61 / 2.77 / 1.54 s (26–37 % lower per pair)
centre moments max abs diff vs PR 1	–	0.0 (5 runs, bit-identical)
ep_history flags	S26 / NC15 / F0 / BP10 / E0	identical
PNGs under dataset_*/optimization_*	per-search corner / fit images	only optimization_0/image/data.png per factor

Full evidence: #1642 (comment).

API Changes

No public signature is removed or changed incompatibly. One behaviour change and two additions:

• EP factor searches no longer write per-search visuals (corner plots, model-fit images) and no longer redraw before-fit visuals after the first search of each factor. Set output.visualize_ep_factor_searches: true in general.yaml to restore the previous behaviour. The NSS live-visual path is untouched, and ordinary search.fit (outside EP) is unchanged.
• New config key output.visualize_ep_factor_searches (default false).
• New keyword SearchUpdater(..., visualization_enabled: bool = True) and attribute SearchUpdater.visualization_enabled.

See full details below.

Test Plan

• New test_autofit/graphical/test_ep_factor_search_visuals.py (3 tests, all red on PR 1 source):
  • default key: a two-factor EP run with a named DynestyStatic makes 0 visualize and 0 plot_results calls and one visualize_before_fit per factor; graph.info, graph.png, ep_history.csv present;
  • key true: every factor search draws per-search visuals;
  • the visual switches are back to True after optimise.
• test_updater.py::test__visualize__disabled_skips_all.
• Full suite, run serially (pytest test_autofit -p no:xdist): 2885 passed, 2 skipped.
• EP harness: moments bit-identical to PR 1, per-search wall 26–37 % lower in load-matched pairs.
• Downstream impact (iii) none: visualize_ep_factor_searches, visualization_enabled, _visualize_fit, _visualize_before_fit have no hits in autofit_workspace*, HowToFit, PyAutoArray, PyAutoGalaxy or PyAutoLens.

Full API Changes (for automation & release notes)

Added

• Config general.yaml -> output -> visualize_ep_factor_searches (default false).
• SearchUpdater.__init__(..., visualization_enabled: bool = True) and the public attribute SearchUpdater.visualization_enabled.
• NonLinearSearch._visualize_fit, NonLinearSearch._visualize_before_fit (private class attributes, default True).

Changed Behaviour

• NonLinearSearch.optimise(...) (EP factor search entry point): sets _visualize_fit = visualize_ep_factor_searches and _visualize_before_fit = visualize_ep_factor_searches or number == 0 around fit, resetting both to True in a finally.
• NonLinearSearch.pre_fit_output: analysis.visualize_before_fit now also requires _visualize_before_fit.
• SearchUpdater.visualize: returns immediately when visualization_enabled is False.
• NonLinearSearch._updater: re-applies _visualize_fit to the cached updater on every access.

Migration

• None required. To keep per-factor EP images, set visualize_ep_factor_searches: true under output: in your general.yaml.

Heart RED override (development only)

This PR is opened under the PyAutoBrain/AUTONOMY.md "Human override for Heart RED (development only)". It is a development-shipping override only. It does not claim to fix Heart, and Heart remains RED for release purposes. Merge still needs a separate explicit human /prm with every required check green.

• Authorisation (live human, 2026-09-24, Claude Code CLI session): asked "Does your override extend to opening PR 2 now as a stacked PR (base = PR 1 branch, retargeted to main after PR 1 merges), and PR 3 the same way once it is green?", the human selected "Yes, same override for PR 2 and PR 3".
• Exact current RED reasons (pyauto-heart readiness --json, verdict red, score 45, ts 2026-09-24T15:11:12.973142+00:00):
  1. release validation FAILED (stage integrate)
  2. workspace validation not passing (4 failed, cloud#35579888156: autolens notebooks/cluster/modeling.ipynb, autolens notebooks/weak/a2744.ipynb, autolens scripts/cluster/modeling.py, +1 more)
  3. manifest drift: hub organism blurb (organs present) — 7 mismatch(es) vs PyAutoMind/repos.yaml
• Branch gates passed: full serial test_autofit 2885 passed / 2 skipped on head 1cd4a5b; the new EP-visuals tests are red on PR 1 source; EP harness moments bit-identical to PR 1; downstream workspace impact (iii) none.
• Permitted scope: commit, push, and opening this stacked pending-release PR. No merge, release or release rehearsal, and no bypassing of tests, checks or branch protection.

Generated by the PyAutoLabs agent workflow.

🤖 Generated with Claude Code

https://claude.ai/code/session_01EQdWrHM9BUx1q2zZRy6Y7N