Skip to content

feat: PYAUTO_SMALL_DATASETS + canonical auto-sim for the cluster scripts#250

Merged
Jammy2211 merged 1 commit into
mainfrom
feature/cluster-small-datasets
Jul 9, 2026
Merged

feat: PYAUTO_SMALL_DATASETS + canonical auto-sim for the cluster scripts#250
Jammy2211 merged 1 commit into
mainfrom
feature/cluster-small-datasets

Conversation

@Jammy2211

Copy link
Copy Markdown
Collaborator

Summary

PYAUTO_SMALL_DATASETS + canonical auto-simulation for the cluster path (issue #249; the final filed item of the cluster home straight). The diagnosis was good news: the libraries' small-mode hooks (PointSolver fixed-pair short-circuit, 15×15 grid caps) already make the cluster scripts fast — the historical "test mode breaks it" parking simply predates them, and nothing ever wired the scripts in.

  • scripts/cluster/{modeling,start_here}.py — hand-rolled three-file existence checks replaced by the canonical al.util.dataset.should_simulate guard (matching imaging, and the weak twin shipping via PyAutoLens#581), which also carries small-mode's delete-and-regenerate semantic so smoke datasets actually get built small.
  • scripts/cluster/lenstool/data.py — the 96 MB RELICS mosaic download/cutout (a visualization-only product) is skipped under PYAUTO_SMALL_DATASETS=1; the Mahler model files (~85 KB) still download so the modeling products are complete.
  • scripts/cluster/lenstool/modeling.py — the production-scale refit now runs only under its explicit LENSTOOL_EXAMPLE_RUN_FIT=1 opt-in, never PYAUTO_TEST_MODE (the sweep env put the 72-parameter factor-graph search into reduced-iterations mode — a 600 s+ hang; the structural pass is the model composition + model.info, which always runs).

Evidence (fresh dataset regeneration, exact sweep env from config/build/env_vars.yaml defaults)

Script Result
cluster/start_here.py exit 0 in 24 s
cluster/modeling.py exit 0 in 26 s
cluster/lenstool/data.py exit 0 in 2 s (mosaic skipped)
cluster/lenstool/modeling.py exit 0 in 175 s (verification in full; the 72-param search gated behind its explicit opt-in)

Companion PyAutoBuild PR removes the two no_run.yaml cluster lines — the task's success criterion.

Scripts Changed

  • Modified: scripts/cluster/modeling.py, scripts/cluster/start_here.py, scripts/cluster/lenstool/data.py, scripts/cluster/lenstool/modeling.py (+ navigator catalogue regen).

Merge coordination

Three open branches touch the navigator catalogue (this, point-pairing-policies #248, and the weak thread) — whichever merges last regenerates llms-full.txt/workspace_index.json on main (two commands, noted in all three).

Validation checklist (--auto run — in-session directives)

  • Effective level: supervised (cap: feature → supervised); "continue --auto"; Heart-ack carried in-session
  • Human: merge with the companion PyAutoBuild PR

🤖 Generated with Claude Code

…er scripts

cluster/{modeling,start_here}.py: manual three-file existence guards ->
al.util.dataset.should_simulate (imaging convention; carries small-mode
delete-and-regenerate). lenstool/data.py: 96 MB mosaic/cutout leg skipped
under PYAUTO_SMALL_DATASETS=1 (visualization-only product). lenstool/
modeling.py: the 72-param refit runs only under LENSTOOL_EXAMPLE_RUN_FIT=1,
never PYAUTO_TEST_MODE (sweep env put the factor-graph search into reduced-
iterations mode -> 600s+ hang; structural pass always runs). Evidence under
the exact sweep env, fresh regen: start_here 24s, modeling 26s, lenstool
data 2s, lenstool modeling 175s (0.068" parity intact).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PUuWXiS23FvmfQPLvMNjeM
@Jammy2211 Jammy2211 added the pending-release PR queued for the next release build label Jul 9, 2026
@Jammy2211 Jammy2211 merged commit 1229235 into main Jul 9, 2026
8 checks passed
Jammy2211 added a commit that referenced this pull request Jul 9, 2026
docs: Lenstool example on the canonical CSV API end to end (stacked on #250)
@Jammy2211 Jammy2211 deleted the feature/cluster-small-datasets branch July 9, 2026 13:49
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