Skip to content

docs: cluster-pointsolver spurious-core check-in (PyAutoLens#800) #73

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
Jammy2211 merged 2 commits into from
Oct 23, 2026
Merged

docs: cluster-pointsolver spurious-core check-in (PyAutoLens#800) #73

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
d523c37
docs: cluster-pointsolver spurious-core check-in (PyAutoLens#799 → #800)
Jammy2211 Oct 23, 2026
fc81800
docs: spurious-core filter on by default (human decision) + side issues
Jammy2211 Oct 23, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 8 additions & 0 deletions tasks/cluster_pointsolver_speed.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,6 +40,14 @@ not refresh measurements or authorize compute. Historical task text is retained.
- Human decision TAKEN 2026-10-23 (chat): keep the 1.0″ starting grid, plus the validation check as mitigation. See the Decision section below.
- Next bounded task: spurious-core candidates (`draft/feature/autolens/point_solver_spurious_core_candidates.md`, start_dev in progress), then the workspace validation check.
- Review date does not refresh measurement dates. This check-in authorizes no compute, default change or merge. [Full dated check-in](../checkins/2026-10-23.md).
- Spurious-core candidates ISSUED and SHIPPED 2026-10-23: [PyAutoLens#799](https://github.com/PyAutoLabs/PyAutoLens/issues/799) → [PyAutoLens#800](https://github.com/PyAutoLabs/PyAutoLens/pull/800) (OPEN, `pending-release`, awaiting human `/prm`). It adds `residual_threshold`, which rejects candidates with |β(θ̂) − β| > k · `pixel_scale_precision`. **Human decision 2026-10-23: on by default at k = 10** (the problem is common on cluster models; `None` disables it).
- Mechanism: the host is a near-critical saddle at the `r_core = 0` members, so the core points have |μ| 0.1–0.3 and the magnification filter cannot remove them.
- Evidence (laptop CPU fp64, diagnostic): true candidates reach at most 3.58× the precision (viz), 2.21× (`simple`) and 1.075× (1c/1d near-cusp); the core points are at least 38.5×. At k = 10 all 4 viz core points are removed and nothing else changes in 238 cases, and the 1c/1d rows are bit-identical.
- The rejected rule, a triangle-area consistency check, drops true near-cusp candidates.
- Not tested: A2744. At precision 0.01 the cut (0.1″) is above the 0.04″ residuals.
- Default-on checks against the branch: 952 library tests pass; workspace smoke passes (autolens 41 + 2 notebooks, workspace_test 37, euclid 9, HowToLens 50); 6 real-solve workspace_test point-source scripts pass. The filter costs one extra trace of the padded candidates per solve.
- Side issues filed: [PyAutoGalaxy#658](https://github.com/PyAutoLabs/PyAutoGalaxy/issues/658) (`MassSheet` fails under JAX); [autolens_profiling#445](https://github.com/PyAutoLabs/autolens_profiling/issues/445) (the stale auto-simulated `dataset/cluster/simple` is never regenerated).
- Next: the workspace validation-check draft, then a workspace_test CI witness for the default-on `residual_threshold` (phase 2). One task at a time.

## Knowledge (2026-10-22): phase 1, cluster-scale robust PointSolver settings

Expand Down
Loading