You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
refactor: import configs only from sbi.neural_nets, remove build_zuko_cnf, merge z-score embedding helpers - #2162
Removes leftovers of the NN builder refactor before 0.28. The four commits are independent, so each can be dropped on its own.
Configs only from sbi.neural_nets. The per-model configs were also re-exported from sbi.neural_nets.net_builders. The docs, the API reference and sbi itself only use sbi.neural_nets, and the second path was never released. net_builders now exports only build functions.
Remove build_zuko_cnf. CNF left the registry during Zuko density estimators (#1088) #1116, but its builder stayed. posterior_nn("zuko_cnf") raises NotImplementedError at build time, and nothing calls the builder. build_zuko_flow("CNF", ...) builds the same network (same weights and log-probs) and stays tested.
Make build_context internal.ZScoreConfig, ZScoreStats, BuildContext and compute_z_score_stats have no consumer yet. Following Decision 10 of the NN builder design, they stay private until the planned z-score pre-pass uses them: the module becomes sbi.neural_nets._build_context. It was never released. @satwiksps, this settles the "unused seam" item of your post-GSoC plan. If you'd rather remove the module, that is one commit.
One z-score embedding helper. Seven places prepended a z-scoring layer to an embedding net with the same code. build_z_scored_embedding_net moves unchanged to sbi.utils.sbiutils, because the restriction estimator can't import from sbi.neural_nets, and every builder now calls it. Each call stays where the old code was. The old classifier import path and build_input_layer keep working.
Release note:build_zuko_cnf (shipped since v0.23) is removed; use sbi.neural_nets.net_builders.flow.build_zuko_flow("CNF", ...). No other behaviour changes.
Does this close any issues?
No issue. It is cleanup from the NN builder refactor (#1872 to #1997).
Anything else we should know?
Same networks as before: for 19 builder entry points × 6 z-score settings × 2 embedding nets (228 cases), the module structure, state_dict, an output and any error message are bit-identical to main. Three planted bugs were caught by this comparison: ignoring structured, wrapping before get_numel, and copying the embedding net.
Tests: the fast suite passes. The slow tests of the 23 touched test files pass too; their xfails also xfail on main.
Only behavioural difference: in the categorical builder, an invalid z_score_y now raises the same ValueError after get_numel instead of before it.
The per-model config classes were re-exported from both
`sbi.neural_nets` and `sbi.neural_nets.net_builders`. The docs, the API
reference and sbi itself only use `sbi.neural_nets`, and the second
path was never released. `net_builders` now exports only the build
functions.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
CNF was dropped from the model registry when the Zuko flows were added
in sbi-dev#1116, but its builder stayed. `posterior_nn("zuko_cnf")` raises
`NotImplementedError` when it builds, and nothing in sbi calls the
builder. `build_zuko_flow("CNF", ...)` builds the same network and stays
tested.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
`ZScoreConfig`, `ZScoreStats`, `BuildContext` and `compute_z_score_stats`
have no consumer yet. As decided in the NN builder design (Decision 10),
they stay internal until the planned z-score pre-pass uses them. Rename
the module to `sbi.neural_nets._build_context`, so 0.28 does not ship
them as public API. The module was never released.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Seven places prepended a z-scoring layer to an embedding net with the
same code: `build_z_scored_embedding_net` in the classifier builders,
`_prepare_y_embedding` in the flow builders, `build_input_layer` in the
restriction estimator, and inline copies in the MDN, categorical, mixed
and vector-field builders.
Move `build_z_scored_embedding_net` unchanged into `sbi.utils.sbiutils`,
next to `standardizing_net`, and use it everywhere. It lives in
`sbi.utils` because the restriction estimator cannot import from
`sbi.neural_nets` without a circular import. The classifier module
imports it back, and `build_input_layer` keeps its signature, so both
released names keep working. Every call keeps its position, so the
embedding net still runs at the same points of each build.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
No actionable comments were generated in the recent review.
Walkthrough
The pull request adds a shared helper for z-score embedding setup and updates builder and restriction-estimator call sites to use it. It also removes selected configuration re-exports and the Zuko CNF builder.
Changes: Z-score embedding setup
Layer / File(s)
Summary
Shared helper and validation sbi/utils/sbiutils.py, sbi/neural_nets/net_builders/classifier.py, tests/sbiutils_test.py, sbi/neural_nets/_build_context.py
build_z_scored_embedding_net now parses z-score modes and composes a standardizing network with the supplied embedding network when enabled. Tests cover supported modes and compatibility imports. The build-context documentation now describes the types as internal and not yet used by builders.
The intended API cleanup has no identified break in repository workflows. The PR is mergeable after normal checks.
Pre-merge checks: 5
Check name
Status
Explanation
Title check
The title names the main changes: config exports, removal of build_zuko_cnf, and consolidation of z-score embedding helpers. It is longer than preferred but remains clear and specific.
Description check
The description follows the template and explains the changes, issue status, test results, compatibility details, and AI use. It completes the checklist and explains why the test-timing item does not …
Docstring Coverage
Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff with cyclomatic complexity of at least 2. Documentat…
Linked Issues check
Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check
Check skipped because no linked issues were found for this pull request.
Full details: Description check
Explanation
The description follows the template and explains the changes, issue status, test results, compatibility details, and AI use. It completes the checklist and explains why the test-timing item does not apply.
Full details: Docstring Coverage
Explanation
Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff with cyclomatic complexity of at least 2. Documentation-only edits and deletions within surviving functions are included; unchanged diff context and fully deleted functions are excluded. Analyzed 12 functions across 11 files.
Finishing Touches
Generate unit tests (beta)
Create a new PR
Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.
File: No configuration file from this repository applied
Commit: Unavailable
Layers: Organization UI
Review profile: CHILL
Plan: Advanced
Run ID: 134d7b46-fe75-4901-9c22-7db0f348f29e
Commits
Reviewing files that changed from the base of the PR and between c36f471 and dc1f46e.
Files selected for processing: 12
sbi/neural_nets/_build_context.py
sbi/neural_nets/net_builders/__init__.py
sbi/neural_nets/net_builders/categorial.py
sbi/neural_nets/net_builders/classifier.py
sbi/neural_nets/net_builders/flow.py
sbi/neural_nets/net_builders/mdn.py
sbi/neural_nets/net_builders/mixed_nets.py
sbi/neural_nets/net_builders/vector_field_nets.py
sbi/utils/restriction_estimator.py
sbi/utils/sbiutils.py
tests/build_context_test.py
tests/sbiutils_test.py
Files with no reviewable changes: 1
sbi/neural_nets/net_builders/init.py
Included review availability: This review used your included allowance. Your plan provides up to 4 included reviews per hour; 3 remain after this review.
❌ Patch coverage is 94.44444% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 90.59%. Comparing base (c36f471) to head (dc1f46e).
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
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.
What does this PR do?
Removes leftovers of the NN builder refactor before 0.28. The four commits are independent, so each can be dropped on its own.
sbi.neural_nets. The per-model configs were also re-exported fromsbi.neural_nets.net_builders. The docs, the API reference and sbi itself only usesbi.neural_nets, and the second path was never released.net_buildersnow exports only build functions.build_zuko_cnf. CNF left the registry during Zuko density estimators (#1088) #1116, but its builder stayed.posterior_nn("zuko_cnf")raisesNotImplementedErrorat build time, and nothing calls the builder.build_zuko_flow("CNF", ...)builds the same network (same weights and log-probs) and stays tested.build_contextinternal.ZScoreConfig,ZScoreStats,BuildContextandcompute_z_score_statshave no consumer yet. Following Decision 10 of the NN builder design, they stay private until the planned z-score pre-pass uses them: the module becomessbi.neural_nets._build_context. It was never released. @satwiksps, this settles the "unused seam" item of your post-GSoC plan. If you'd rather remove the module, that is one commit.build_z_scored_embedding_netmoves unchanged tosbi.utils.sbiutils, because the restriction estimator can't import fromsbi.neural_nets, and every builder now calls it. Each call stays where the old code was. The old classifier import path andbuild_input_layerkeep working.Release note:
build_zuko_cnf(shipped since v0.23) is removed; usesbi.neural_nets.net_builders.flow.build_zuko_flow("CNF", ...). No other behaviour changes.Does this close any issues?
No issue. It is cleanup from the NN builder refactor (#1872 to #1997).
Anything else we should know?
main. Three planted bugs were caught by this comparison: ignoringstructured, wrapping beforeget_numel, and copying the embedding net.main.z_score_ynow raises the sameValueErrorafterget_numelinstead of before it.build_zuko_cnf(keep the deletion). Simformer #1621 adds two names to an import block this PR deletes (re-add them as their own import). This PR merges cleanly with fix: embedding nets end in a linear layer, and two bugs in network building #2118, fix: potentials follow the caller's grad mode by default #2127, docs: fix stale method names, crashing examples and a default in docstrings #2145, fix: keep Standardize mean and std only as buffers, so .to() moves them #2155 and feat: parallel simulation with simulate_from_theta and parallelize_simulator #1738.Checklist
uv run pytest -n auto -m "not slow and not gpu"passes.uv run pre-commit run --all-filespasses (ruff and formatting).uv run pyright sbipasses.pytest.mark.slow. (Not applicable: the new tests take milliseconds.)🤖 Generated with Claude Code