Repository navigation
jax: register FitImaging, Tracer, DatasetModel as pytrees on use_jax path - #445
Merged
Merged
Conversation
…path AnalysisImaging.fit_from now calls _register_fit_imaging_pytrees on the JAX path, so jax.jit(analysis.fit_from)(instance) returns a real FitImaging with jax.Array leaves instead of raising TypeError at the JIT output boundary. dataset, adapt_images, settings, and cosmology ride as aux_data (per-analysis constants); tracer, dataset_model, and nested autoarray wrappers flow through as dynamic children. Updates CLAUDE.md to document the pytree-wrapper-return pattern alongside the existing if-xp-is-np raw-array pattern. Issue: #444
This was referenced Apr 19, 2026
Collaborator
Author
|
Workspace PR: PyAutoLabs/autolens_workspace_test#32 |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Path A for the
fit-imaging-pytreetask (issue #444).AnalysisImaging.fit_fromnow registersFitImaging,Tracer, andDatasetModelas JAX pytree nodes on the first call when_use_jaxis true. With this in place,jax.jit(analysis.fit_from)(instance)returns a realFitImagingwithjax.Arrayleaves — previously it raisedTypeError: ... is not a valid JAX typeat the JIT output boundary.dataset,adapt_images,settings, andcosmologyare pushed throughaux_data(per-analysis constants, closed over the JIT boundary) via theno_flattenoption on the genericregister_instance_pytreehelper added in the companion PyAutoArray PR. Onlytracer,dataset_model, and their nested autoarray wrappers flow through as dynamic children.CLAUDE.mdis updated: the "autoarray types cannot be returned fromjax.jit" paragraph now documents two coexisting patterns (the existingif xp is np:raw-array guard, and the new pytree-registered wrapper return).PoC:
autolens_workspace_test/scripts/jax_likelihood_functions/imaging/mge_pytree.py—jax.jit(analysis.fit_from)(instance).log_likelihoodmatches the NumPy path to ~1e-8 relative error for the MGE parametric lens + MGE source model.Depends on: PyAutoLabs/PyAutoArray#288
Issue: #444
API Changes
One private addition, no public removals/renames:
AnalysisImaging._register_fit_imaging_pytrees()— registersFitImaging,Tracer,DatasetModelas JAX pytree nodes. Called fromfit_fromwhenself._use_jaxis true. Idempotent.jax.jit(analysis.fit_from)(instance)now succeeds on the JAX path. Previously it raised at the JIT output boundary.No removals, renames, or public signature changes.
See full details below.
Test Plan
pytest test_autolens/— 246 passedpytest test_autoarray/— 736 passedpytest test_autogalaxy/— 836 passedpytest test_autofit/— 1232 passedautolens_workspace_test/scripts/jax_likelihood_functions/imaging/mge_pytree.pyreturns-86629.349379496underjax.jit, matching the NumPy scalar to ~1e-8 relFull API Changes (for automation & release notes)
Added
autolens.imaging.model.analysis.AnalysisImaging._register_fit_imaging_pytrees()(staticmethod) — registersFitImaging,Tracer, andDatasetModelas JAX pytree nodes via the genericautoarray.abstract_ndarray.register_instance_pytreehelper.dataset,adapt_images,settings, andcosmologyride asaux_data. Called once at the top offit_fromwhenself._use_jaxis true. Idempotent.Changed Behaviour
AnalysisImaging.fit_from(instance)— when constructed withuse_jax=True, the returnedFitImagingcan now be returned from ajax.jit-compiled function. Previouslyjax.jit(analysis.fit_from)(instance)raisedTypeError: ... FitImaging ... is not a valid JAX type.Migration
None required. Pure NumPy callers see no change. JAX callers who were previously working around the output-boundary error can now return
FitImagingdirectly from jit'd code.🤖 Generated with Claude Code