Skip to content

plot_coverage_followups.md

Latest commit

 

History

History
122 lines (92 loc) · 5.39 KB

plot_coverage_followups.md

File metadata and controls

122 lines (92 loc) · 5.39 KB
Raw

plot coverage — follow-ups deferred from plot-coverage-gaps

Four items deliberately scoped out of plot-coverage-gaps (PyAutoLens#667 / #668, autolens_workspace#404, autogalaxy_workspace#191). Each is independent; do not bulk-issue them as a series.

Recorded 2026-07-30, at that task's PR-open.

1. Demo subplot_fit_interferometer_combined

PyAutoLens#668 exported aplt.subplot_fit_interferometer_combined for API symmetry — imaging's subplot_fit_combined was exported, the interferometer equivalent was not — but nothing demonstrates it.

It takes a fit_list, so its home is a multi-dataset interferometer example, not the single-fit scripts/interferometer/plot.py. Passing [fit, fit] there would teach a wrong idiom (the same reasoning that kept the *_x1_plane pair out of that task).

Candidate home: autolens_workspace/scripts/multi/ — multi/plot.py already demonstrates the imaging combined subplots (subplot_fit_combined, subplot_fit_combined_log10), and multi/features/imaging_and_interferometer/ provides a real multi-interferometer fit context.

  • @autolens_workspace

2. Demo subplot_ellipse_errors

ag.plot.subplot_ellipse_errors is demonstrated nowhere in autogalaxy_workspace.

It takes fit_pdf_list: List[List[FitEllipse]] — outer list one entry per posterior sample, inner list one FitEllipse per ellipse. A standalone plot.py runs no search, so this needs a real ellipse model-fit to produce genuine samples. Do not fake it by perturbing parameters: the figure's whole point is the inference-derived error region.

Likely home: an ellipse results/fit.py-adjacent script that already has a Result, rather than scripts/ellipse/plot.py.

  • @autogalaxy_workspace

3. docs/api/plot.rst omits already-exported symbols

PyAutoLens#668 added its two new exports to docs/api/plot.rst, but the file was already missing several symbols that autolens.plot exports:

  • subplot_imaging_dataset
  • subplot_imaging_dataset_list
  • fits_imaging
  • fits_interferometer
  • subplot_fit_interferometer_tracer
  • subplot_interferometer_dirty_images

There is no "Dataset Subplots" heading in the rst at all — the dataset-level functions have no home in it. Check the autogalaxy equivalent for the same class of drift, and consider whether /audit_docs should cover plot.rst autosummary blocks against the live aplt namespace.

  • @PyAutoLens (and probably @PyAutoGalaxy)

4. autolens.plot shadows two of its own interferometer functions — RESOLVED

SHIPPED 2026-07-30 as PyAutoLens#670 (339867e2c). Resolved additively: subplot_fit_interferometer_dirty_images now exports autolens's own function, matching the existing subplot_fit_real_space / subplot_fit_interferometer_real_space convention. subplot_fit_dirty_images was deliberately left bound to autogalaxy's version — the signatures diverge (residuals_symmetric_cmap vs image_plane_lines), so rebinding would be a behaviour change rather than an additive export.

Confirmed the autolens version is materially better for lensing, not a duplicate: with image_plane_lines=None it auto-derives the tracer's critical curves from the fit (_compute_critical_curve_lines) and overlays them on the dirty model image. Verified functionally on a real FitInterferometer — 1 curve derived — not just signature-checked.

Remaining sub-item: autolens_workspace still calls the autogalaxy-bound aplt.subplot_fit_dirty_images at 11 sites — scripts/interferometer/{fit,modeling,plot}.py, four scripts/interferometer/features/*/fit.py, and scripts/multi/features/imaging_and_interferometer/modeling.py. All pass fit=fit only (none pass residuals_symmetric_cmap), so all 11 can be switched to the new name and would gain critical-curve overlays for free. Deferred because it is a visual change across 11 lens examples and deserves its own review pass.

Original finding

aplt.subplot_fit_dirty_images and aplt.subplot_fit_real_space resolve to the autogalaxy implementations inside autolens.plot:

>>> aplt.subplot_fit_dirty_images.__module__
'autogalaxy.interferometer.plot.fit_interferometer_plots'

autolens has its own versions in autolens/interferometer/plot/fit_interferometer_plots.py which accept lensing-specific image_plane_lines / image_plane_line_colors / source_plane_lines arguments the autogalaxy versions do not. Lens users calling aplt.subplot_fit_dirty_images therefore silently get the galaxy version and cannot pass critical-curve/caustic overlays.

Note subplot_fit_interferometer_real_space is separately exported as autolens's own, so subplot_fit_real_space (AG) and subplot_fit_interferometer_real_space (AL) currently coexist under different names — confusing but not lossy. subplot_fit_dirty_images is the lossy one: autolens's version is reachable under no exported name.

This was kept out of #668 deliberately: rebinding an existing exported name is a behaviour change, not an additive export. Decide whether to rebind (and accept the behaviour change), add a distinct subplot_fit_interferometer_dirty_images name, or leave it — then sweep the workspace calls accordingly.

  • @PyAutoLens
  • @autolens_workspace (calls aplt.subplot_fit_dirty_images in scripts/interferometer/plot.py, scripts/interferometer/modeling.py and scripts/multi/features/imaging_and_interferometer/modeling.py)