From b9cedf3c2f91479a768d0ca70fde317ef30b2293 Mon Sep 17 00:00:00 2001 From: Claude Date: Tue, 18 Aug 2026 02:24:09 +0000 Subject: [PATCH] Shelve superseded group4 MGE search benchmark MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The group4 MGE search benchmark (parked.md :: group4-mge-search-benchmark, autolens_profiling#82) is superseded by the 2026-08-17 inference-methods programme (active/inference_programme_ledger.md, autolens_profiling#134), which now owns sampler-benchmark direction. Its first half already shipped (code + first GPU results merged as autolens_profiling#83); the remaining gradient-family sweep is not carried forward as scoped. Like the sampler wave, it showed on the dashboard twice — once under In flight (the prompt still sat in active/) and once under Parked. - move active/research_profiling_experiment_in_the_autolens_pr.md -> complete/archive/shelved/, with a SUPERSEDED header preserving the full parked context (remaining sweep + GPU env, post-#84 script paths, sweep driver invocation, compile-cost knobs) that the deleted parked.md entry carried - drop the parked.md entry - regenerate dashboard.md (in flight 9 -> 8, parked 3 -> 2) Dated completion records that name the task (complete/2026/08/active-prompt-orphan-triage.md, complete/2026/07/profiling-mirror-taxonomy.md) are left as history; the restructure the latter waited on has since landed (#84). lifecycle.py check: OK Co-Authored-By: Claude Opus 5 Claude-Session: https://claude.ai/code/session_016X8zrRQuau8VoGwx3i45tf --- ...profiling_experiment_in_the_autolens_pr.md | 14 ------- ...profiling_experiment_in_the_autolens_pr.md | 40 +++++++++++++++++++ dashboard.md | 8 ++-- parked.md | 7 ---- 4 files changed, 43 insertions(+), 26 deletions(-) delete mode 100644 active/research_profiling_experiment_in_the_autolens_pr.md create mode 100644 complete/archive/shelved/research_profiling_experiment_in_the_autolens_pr.md diff --git a/active/research_profiling_experiment_in_the_autolens_pr.md b/active/research_profiling_experiment_in_the_autolens_pr.md deleted file mode 100644 index 62d1925a..00000000 --- a/active/research_profiling_experiment_in_the_autolens_pr.md +++ /dev/null @@ -1,14 +0,0 @@ -# Research profiling experiment in the autolens_profiling repo - -Type: research -Target: workspaces -Repos: -- autolens_profiling -Difficulty: hard -Autonomy: safe -Priority: normal -Status: formalised - -Research profiling experiment in the autolens_profiling repo. We currently have autolens_profiling examples that run JAX gradient max-likelihood optimizers on a single lens galaxy with a single MGE source. Extend this to a much higher-dimensional, harder model: 4 lens galaxies + 4 source galaxies. Write a simulator.py that generates the dataset from known input truth, then run the existing JAX gradient optimizers (max-likelihood samplers) alongside Nautilus and record whether any of them scale to this dimensionality and recover the input truth. If none of the optimizers succeed, investigate more careful initialization strategies. This is an exploratory benchmark, not a library change. - - diff --git a/complete/archive/shelved/research_profiling_experiment_in_the_autolens_pr.md b/complete/archive/shelved/research_profiling_experiment_in_the_autolens_pr.md new file mode 100644 index 00000000..4a8d1ccf --- /dev/null +++ b/complete/archive/shelved/research_profiling_experiment_in_the_autolens_pr.md @@ -0,0 +1,40 @@ +# Research profiling experiment in the autolens_profiling repo (group4 MGE search benchmark) — SUPERSEDED 2026-08-18 + +> **SUPERSEDED / SHELVED 2026-08-18.** Removed from the registry (was +> `parked.md :: group4-mge-search-benchmark`, issue +> [autolens_profiling#82](https://github.com/PyAutoLabs/autolens_profiling/issues/82)) +> — superseded by the 2026-08-17 human-approved inference-methods programme +> (`active/inference_programme_ledger.md`, autolens_profiling#134), which now +> owns sampler-benchmark direction. The first half already SHIPPED (code + first +> GPU results merged as autolens_profiling#83); the remaining gradient-family +> sweep is not carried forward as scoped here. +> +> **Preserved parked context (from the deleted `parked.md` entry):** parked +> 2026-07-24 — code + first GPU results MERGED (PR #83), worktree/claim RELEASED. +> *Remaining at park time:* gradient-family sweep +> (prodigy / lion / adabelief / prodigy_autoconv) + Nautilus anchor on the laptop +> GPU (`~/venv/PyAutoGPU`, `JAX_PLATFORM_NAME=cuda JAX_PLATFORMS=cuda,cpu +> XLA_PYTHON_CLIENT_MEM_FRACTION=0.5`, `--config-name local_gpu_fp64`), then +> recovery/walltime aggregation. Warm output preserved in the main checkout under +> `output/searches/`. *Paths (post scripts/// restructure, +> autolens_profiling#84):* group4 cells live at +> `scripts/cluster/searches//mge.py` (samplers: `multi_start_prodigy`, +> `multi_start_lion`, `multi_start_adabelief`, `multi_start_prodigy_autoconv`, +> `nautilus`); run via the sweep driver `scripts/misc/searches/sweep.py` +> (e.g. `--only /group/mge` — the sweep still keys the group cell class +> internally, mapping it to `scripts/cluster/` on disk). If compile is too heavy: +> dial `_GROUP4_MGE_TOTAL_GAUSSIANS` 10->6 and/or `_MULTI_START_N_STARTS` 64->32. + + +Type: research +Target: workspaces +Repos: +- autolens_profiling +Difficulty: hard +Autonomy: safe +Priority: normal +Status: formalised + +Research profiling experiment in the autolens_profiling repo. We currently have autolens_profiling examples that run JAX gradient max-likelihood optimizers on a single lens galaxy with a single MGE source. Extend this to a much higher-dimensional, harder model: 4 lens galaxies + 4 source galaxies. Write a simulator.py that generates the dataset from known input truth, then run the existing JAX gradient optimizers (max-likelihood samplers) alongside Nautilus and record whether any of them scale to this dimensionality and recover the input truth. If none of the optimizers succeed, investigate more careful initialization strategies. This is an exploratory benchmark, not a library change. + + diff --git a/dashboard.md b/dashboard.md index e99cd35f..af95fc09 100644 --- a/dashboard.md +++ b/dashboard.md @@ -8,8 +8,8 @@ Tasks only — the organism's health lives with the Heart (`/health`), not here. | Where | Count | |-------|------:| -| [In flight](#in-flight) (`active/`) | 9 | -| [Parked](#parked) (`parked.md`) | 3 | +| [In flight](#in-flight) (`active/`) | 8 | +| [Parked](#parked) (`parked.md`) | 2 | | [Planned](#planned) (`planned.md`) | 7 | | [Backlog](#backlog) (`draft/`) | 139 | @@ -53,7 +53,6 @@ Issued — each has an open GitHub issue and usually a branch. The full record f - [Find what kills MGE multi-start lanes — it is not](active/mge_lane_death.md) — [issue #128](https://github.com/PyAutoLabs/autolens_profiling/issues/128) — CAUSE FOUND 2026-08-15 (cloud CPU session) — written up on autolens_profiling#128. Remaining: GPU/float64/multi-seed… - [PositionsLH penalty accumulation returns 2x the LAST penalty, discarding the](active/positions_lh_penalty_accumulation_bug.md) — [issue #699](https://github.com/PyAutoLabs/PyAutoLens/issues/699) — pr-open (https://github.com/PyAutoLabs/PyAutoLens/pull/700) - [PyAutoReduce validation: slacs1430+4105 ACS reduction vs trusted legacy dataset](active/pyautoreduce_slacs1430_acs_comparison.md) -- [Research profiling experiment in the autolens_profiling repo](active/research_profiling_experiment_in_the_autolens_pr.md) — [issue #82](https://github.com/PyAutoLabs/autolens_profiling/issues/82) - [Tenant-firewall drift: clear the 9 right + gate recurrence (Aug](active/tenant-firewall-drift-aug.md) — [issue #198](https://github.com/PyAutoLabs/PyAutoMind/issues/198) — MERGED 2026-08-17T23:13Z — all four PRs on the canonical mains, each green at merge (order held: PyAutoMind#199 → PyAutoBrain#229… - [Reconstructing a stored sample raises through `ignore_assertions=True`](active/to_instance_guard_gap.md) — [issue #1486](https://github.com/PyAutoLabs/PyAutoFit/issues/1486) — library-dev — WORKSPACE HALF SHIPPED; the PyAutoFit hardening (#1486) is what remains - [Version-stamp sync to 2026.8.17.1 + release-sed guards](active/version_stamp_sync_and_release_sed_guards.md) — [issue #235](https://github.com/PyAutoLabs/PyAutoHands/issues/235) — pr-open (https://github.com/PyAutoLabs/PyAutoLens/pull/700) @@ -63,9 +62,8 @@ Issued — each has an open GitHub issue and usually a branch. The full record f Started or scoped, not currently in flight — resume by moving the row back to `active.md`. Full detail in [`parked.md`](parked.md).
-3 task(s) +2 task(s) -- **group4-mge-search-benchmark** — [issue #82](https://github.com/PyAutoLabs/autolens_profiling/issues/82) - **matplotlib-inline-standalones** - **pyautoreduce-slacs1430-acs-comparison** diff --git a/parked.md b/parked.md index 73a25be7..c35f59bf 100644 --- a/parked.md +++ b/parked.md @@ -6,13 +6,6 @@ unindexed worktrees or stashes. Move an entry back to `active.md` (or to `planned.md` if re-scoping is needed) when work resumes; on shipping, write the dated `complete///.md` record instead. -## group4-mge-search-benchmark -- issue: https://github.com/PyAutoLabs/autolens_profiling/issues/82 -- prompt: active/research_profiling_experiment_in_the_autolens_pr.md -- parked: 2026-07-24 — code + first GPU results MERGED (PR #83); worktree/claim RELEASED -- remaining: gradient-family sweep (prodigy/lion/adabelief/prodigy_autoconv) + Nautilus anchor on laptop GPU (~/venv/PyAutoGPU, JAX_PLATFORM_NAME=cuda JAX_PLATFORMS=cuda,cpu XLA_PYTHON_CLIENT_MEM_FRACTION=0.5, --config-name local_gpu_fp64), then recovery/walltime aggregation. Warm output preserved in main checkout output/searches/. -- note: NEW PATHS (scripts/// restructure LANDED, autolens_profiling#84) — group4 cells now live at scripts/cluster/searches//mge.py (samplers: multi_start_prodigy, multi_start_lion, multi_start_adabelief, multi_start_prodigy_autoconv, nautilus); run them via the sweep driver scripts/misc/searches/sweep.py (e.g. `--only /group/mge` — sweep still keys the group cell class internally, mapping it to scripts/cluster/ on disk). If compile still too heavy: dial _GROUP4_MGE_TOTAL_GAUSSIANS 10->6 and/or _MULTI_START_N_STARTS 64->32. - ## matplotlib-inline-standalones - prompt: active/matplotlib_inline_standalones.md - parked: 2026-08-08 — surfaced by the orphaned-prompt triage; VERIFIED INCOMPLETE, not shipped