Honour PYAUTO_TEST_MODE in LOSSampler to fix los_halos simulator timeouts

[Jammy2211]

Summary

The los_halos workspace simulators (simulator.py and simulator_jax.py) timed out at the 300s cap under the integration runner, even with PYAUTO_TEST_MODE=2. These scripts run no non-linear search, so test mode had nothing to short-circuit. The real cost is in LOSSampler.galaxies_from():

• negative_kappa_from runs a nested adaptive scipy.quad with an inner fsolve (~22.5k fsolve calls per plane) — ~3.8s × 9 planes ≈ 35s, and it is grid-independent, so PYAUTO_SMALL_DATASETS=1 never reduced it.
• The sampler draws ~1100 halos per realisation, blowing up downstream ray-tracing and per-galaxy plotting. simulator_jax.py calls galaxies_from 9× (1 + a batch of 8) ≈ 290s of sampling alone.

This change makes LOSSampler honour PYAUTO_TEST_MODE: it caps the halo population to at most 3 per plane and calls negative_kappa_from with quad_limit=1, quad_epsrel=0.1. The expected scipy IntegrationWarning from the low limit is suppressed only in test mode. With test mode off, behaviour is unchanged (signature unchanged).

Verified under the real workspace integration env (PYAUTO_TEST_MODE=2 PYAUTO_SMALL_DATASETS=1 PYAUTO_FAST_PLOTS=1 PYAUTO_DISABLE_JAX=1):

Script	Before (main)	After
simulator.py	timeout 320s (exit 124), 1100 halos	28s, 27 halos
simulator_jax.py	timing out, 1100 halos	45s, 27 halos

Non-test-mode output is bit-identical to main: still 1100 halos with full-accuracy kappa values to 10 decimal places.

API Changes

autolens.lens.los.negative_kappa_from gains two optional keyword arguments, quad_limit=50 and quad_epsrel=1.49e-8 (both scipy's own quad defaults), threaded into its inner and outer integrals. Existing callers are unaffected. LOSSampler.galaxies_from now reads autoconf.test_mode.is_test_mode() internally; its signature is unchanged. No removals or renames. See full details below.

Test Plan

• pytest test_autolens/lens/test_los.py (new file): loose-quad result matches a finer reference and stays negative; galaxies_from under PYAUTO_TEST_MODE caps halos to <=3 per plane and emits one negative-kappa MassSheet per plane.
• simulator.py and simulator_jax.py run to exit 0 under the integration env in <50s each.
• Non-test-mode galaxies_from is bit-identical to main (1100 halos, same 9 kappas).

Full API Changes (for automation & release notes)

Added

• autolens.lens.los.negative_kappa_from(..., quad_limit=50, quad_epsrel=1.49e-8) — two optional kwargs controlling the adaptive subdivision limit and relative tolerance of both scipy.integrate.quad calls. Defaults equal scipy's own defaults, so behaviour is unchanged for existing callers.

Changed Behaviour

• autolens.lens.los.LOSSampler.galaxies_from() — when PYAUTO_TEST_MODE is active (autoconf.test_mode.is_test_mode()), caps sampled LOS halos to at most 3 per plane and calls negative_kappa_from with quad_limit=1, quad_epsrel=0.1. The expected scipy IntegrationWarning from the low limit is suppressed only in test mode. With test mode off, behaviour is unchanged (signature unchanged).

Migration

• None required. All changes are backward compatible.

🤖 Generated with Claude Code