Add SP(k) baryonic suppression model by jemme07 · Pull Request #1291 · LSSTDESC/CCL

jemme07 · 2026-05-09T19:41:48Z

Overview

This PR adds the SP(k) baryon-suppression model (Salcido et al. 2023) to pyccl as a new optional baryonic wrapper:

pyccl.BaryonsSPK

Model references:

pyspk reference implementation: https://github.com/jemme07/pyspk
paper: https://doi.org/10.1093/mnras/stad1474

The wrapper keeps CCL-facing wavenumbers in Mpc^-1 and performs the internal conversion to h/Mpc required by pyspk.

What is implemented

1. SP(k) wrapper interface

pyccl/baryons/spk.py implements the SP(k) multiplicative correction:

P_bar(k, a) = P_DMO(k, a) * f_SPk(k, a)

Supported relation modes:
- power_law
- cosmo_power_law
- double_power_law
- binned
Strict validation of required and optional relation parameters.
Clear error if optional dependency is missing:

ModuleNotFoundError: BaryonsSPK requires `pyspk>=2.0.0`

2. CCL integration behavior

Exported through pyccl/baryons/__init__.py and documented under pyccl.baryons API pages.
Public suppression path: boost_factor(cosmo, k, a).
Power-spectrum integration path: _include_baryonic_effects(cosmo, pk).
Calibrated-domain policy:
- boost_factor(...): raises an error if any requested k is above the pyspk calibrated range
- include_baryonic_effects(...): for out-of-range k, uses suppression = 1 (no baryonic correction) to keep Pk2D stable
pyspk warnings are forwarded to CCL warnings and deduplicated per instance.

3. Performance behavior for iterative workflows

Uses cached pyspk.build_sup_model_evaluator(...) objects for repeated calls on the same effective k/h grid.
Cache is bounded and thread-safe (threading.Lock + LRU-style eviction).
Designed to avoid repeated evaluator rebuilds in parameter-inference and MCMC-like loops.

4. Documentation and packaging updates

Added API docs page:
- readthedocs/api/pyccl.baryons.spk.rst
Added install note for optional dependency:
- readthedocs/source/installation.rst
Added changelog note under Unreleased:
- CHANGELOG.md

5. Validation tests and benchmarks

Unit tests in pyccl/tests/test_baryons_spk.py cover:
- smoke tests (scalar/array input)
- direct agreement against pyspk evaluators
- unit-conversion transparency (Mpc^-1 to h/Mpc)
- high-k policy checks
- cosmology integration behavior
- missing dependency path
- warning deduplication behavior
Benchmark tests in benchmarks/test_spk_baryons.py cover:
- reference regression for power_law
- include-vs-boost consistency
- relation-matrix checks across SO and supported modes
Added benchmark reference data:
- benchmarks/data/spk_power_law_fk.txt

Validation run

Focused commands:

pytest -q pyccl/tests/test_baryons_spk.py
pytest -q benchmarks/test_spk_baryons.py

- Introduced `BaryonsSPK` class for SP(k) baryonic suppression using `pyspk>=2.0.0`. - Updated environment.yml to include `pyspk` dependency. - Enhanced installation and API documentation for the new model. - Added unit tests for `BaryonsSPK` functionality and validation.

- Lower default k_min_hmpc from 0.1 to 0.005 h/Mpc so the internal SP(k) grid covers the full physically relevant low-k range and flat left- extrapolation stays near unity (where baryonic effects vanish). - Rewrite _include_baryonic_effects to apply SP(k) only within the calibrated k range [k_min_hmpc, k_max_hmpc] and calibrated redshift range z <= CALIBRATED_Z_MAX (queried from pyspk.constants). CCL's internal Pk2D spline grid extends to k ~ 1500 h/Mpc and a ~ 0.01 (z ~ 99), both far outside pyspk calibration; those cells are left at unity instead of triggering the OOB error policy or returning garbage values. - Relax test_spk_matches_pyspk tolerance from atol=1e-8 to atol=1e-3. CCL evaluates SP(k) on a fixed 128-point internal grid then interpolates, while the test evaluates pyspk directly on the target k array; the resulting interpolation error is O(1e-4), which is well within the ~1% accuracy of the model. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

- Refactored warning handling in BaryonsSPK to use a dedicated function. - Improved error messages for missing required parameters in relation normalization. - Added support for out-of-bounds policy handling in the include_baryonic_effects method. - Updated tests to cover new out-of-bounds behavior and ensure consistency with reference data. - Introduced a new benchmark data file for SP(k) power law reference boost factors.

…tions

…d relation modes for BaryonsSPK

…nd examples

…omparison

- Introduced a context manager to catch warnings when building the model evaluator. - Ensured that warnings are recorded and processed after the evaluator is created. - This change improves the robustness of the BaryonsSPK class by providing better feedback on potential issues during model evaluation.

…ounds handling - Introduced caching for the SP(k) evaluator and its corresponding k-grid to optimize performance. - Simplified the evaluator building process by removing unnecessary conditional checks. - Enhanced out-of-bounds policy handling by removing warnings for internal spline grid usage. - Updated the `boost_factor` and `include_baryonic_effects` methods to utilize the new caching mechanism. - Modified tests to reflect changes in out-of-bounds policy behavior and ensure consistency in baryonic effects calculations.

…and add relation matrix validation

- Updated the calculation of k_min and k_max to be in h-Mpc instead of Mpc. - Adjusted the generation of the k_grid_hmpc to use the new h-Mpc bounds. - Added a check to return a copy of the suppression grid if the mapped k values match the cached k grid.

- Removed k_min_mpc, k_max_mpc, n_k, and out_of_bounds_policy parameters from BaryonsSPK constructor. - Introduced max_evaluator_cache_size to manage evaluator caching. - Simplified evaluator caching logic using OrderedDict. - Updated boost_factor and include_baryonic_effects methods to handle high-k requests more gracefully. - Adjusted tests to reflect changes in out-of-bounds policy handling and removed unnecessary parameters. - Enhanced documentation and comments for clarity.

damonge

Thanks a lot Jaime! Looks great, just a few minor comments/questions.

…aryonsSPK class

jemme07

Thanks for the review @damonge! I addressed all of your comments.

The main change was refactoring the out-of-range handling: instead of the old internal behavior, it now uses explicit user-facing options for both k and z out-of-range cases. I think this makes the behavior clearer and more consistent for users.

Hope this is useful. I’d be happy to make any follow-up tweaks you’d like.

damonge · 2026-05-14T07:32:26Z

Thanks! I think the linter is failing. Maybe check flake8?

coveralls · 2026-05-14T07:37:01Z

Coverage Report for CI Build 25822586014

Warning

Build has drifted: This PR's base is out of sync with its target branch, so coverage data may include unrelated changes.
Quick fix: rebase this PR. Learn more →

Coverage decreased (-0.3%) to 97.205%

Details

Coverage decreased (-0.3%) from the base build.
Patch coverage: 24 uncovered changes across 1 file (193 of 217 lines covered, 88.94%).
No coverage regressions found.

Uncovered Changes

File	Changed	Covered	%
pyccl/baryons/spk.py	216	192	88.89%

Coverage Regressions

No coverage regressions found.

Coverage Stats


Relevant Lines:	7013
Covered Lines:	6817
Line Coverage:	97.21%
Coverage Strength:	0.97 hits per line

💛 - Coveralls

damonge · 2026-05-14T08:55:09Z

Also, a bunch of tests seem to be failing.

damonge · 2026-05-14T10:40:30Z

oh, the failed tests are just to do with coveralls for linux. I can fix those on a different PR, so you can ignore those. Maybe just fix the linting.

… files

jemme07 · 2026-05-14T16:41:49Z

Thanks! I think the linter is failing. Maybe check flake8?

Thanks! I fixed all lint issues. They were all line-length related. I usually run Ruff locally because it is faster, but Flake8 in this repo uses a different line-length setting, so I aligned everything to that.

I also improved some documentation.

oh, the failed tests are just to do with coveralls for linux. I can fix those on a different PR, so you can ignore those. Maybe just fix the linting.

For the Coveralls/Linux failures, I think they are related to parallel coverage reporting in CI. I made a quick fix, so let’s see if that resolves it.

…tion

…nd documentation

…cumentation

jemme07

All donde @damonge . Ready for review again.

jemme07 · 2026-05-15T08:44:54Z

Hey David @damonge, a unit test check failed on macOS (Python 3.11) but it's unrelated to the SP(k) implementation. The failing test (test_caching_lru) checks the caching behaviour but I think in mac the fast calls can vary by more than the allowed tolerance.

I think we can fix this if we check the cache hit counter directly instead of relying on timing. Something like:

hits_before = func.cache_info.hits
timeit_(sigma8=s8_arr[2])
assert func.cache_info.hits == hits_before + 1

Happy to include it here or leave it for a separate PR... whatever you prefer.

jemme07 and others added 27 commits May 5, 2026 18:42

refactor: enhance type annotations and docstrings for BaryonsSPK methods

7c5de10

refactor: enhance type annotations and docstrings for SP(k) test func…

283bf77

…tions

refactor: add module docstrings for SP(k) baryonic suppression files

5d968ac

docs: update module and class docstrings with references and supporte…

2642625

…d relation modes for BaryonsSPK

refactor: improve docstring clarity and formatting for BaryonsSPK class

6d407be

docs: enhance documentation for SP(k) modes with detailed equations a…

b428225

…nd examples

feat: add SP(k) unit-consistent validation notebook for suppression c…

9971544

…omparison

refactor: restructure benchmark tests for SP(k) baryonic suppression …

b0bbe48

…and add relation matrix validation

Updated SPK demo

78eec2d

fix: update warning message path in SP(k) validation notebook

7757be2

fix: simplify description in SP(k) validation notebook

1c1d1cf

fix: enhance description and validation checks in SP(k) notebook

94150a1

CCL unit conversion

24acdcc

fix: improve clarity and consistency in SP(k) validation notebook

24f5184

fix: enhance clarity in BaryonsSPK interpolation and mapping methods

19c37f3

Remove redundant suppression handling in BaryonsSPK class

c537483

refactor: simplify parameter passing in BaryonsSPK initialization

6ba0e7f

fix: update warnings and improve clarity in SP(k) documentation

d852a7a

refactor: implement thread-safe caching in BaryonsSPK class

ab669b4

fix: update warning messages in spk_demo notebook for clarity

f9817ab

damonge reviewed May 11, 2026

View reviewed changes

Comment thread pyccl/baryons/spk.py Outdated

Comment thread pyccl/baryons/spk.py Outdated

Comment thread pyccl/baryons/spk.py Outdated

Comment thread pyccl/baryons/spk.py Outdated

Comment thread pyccl/baryons/spk.py Outdated

Jaime Salcido added 2 commits May 13, 2026 19:48

feat: add out-of-range handling for k and z in BaryonsSPK class

ce12e06

feat: add support for out-of-range handling in BaryonsSPK tests

1771ba2

Jaime Salcido and others added 3 commits May 13, 2026 20:34

feat: enhance documentation for _compute_suppression_grid method in B…

d8fa35d

…aryonsSPK class

feat: improve documentation for relation_params in BaryonsSPK class

86ad815

fix: update warning messages in spk_demo notebook for consistency

a62ea35

jemme07 commented May 13, 2026

View reviewed changes

Comment thread pyccl/baryons/spk.py Outdated

Comment thread pyccl/baryons/spk.py Outdated

Comment thread pyccl/baryons/spk.py Outdated

Comment thread pyccl/baryons/spk.py Outdated

Comment thread pyccl/baryons/spk.py Outdated

jemme07 requested a review from damonge May 13, 2026 20:04

Jaime Salcido and others added 5 commits May 14, 2026 13:47

style: improve code formatting and readability in BaryonsSPK and test…

b3ffd8f

… files

Fix flake8 in notebook

cc311b6

docs: enhance documentation for BaryonsSPK class and update references

79a15e6

feat: add support for parallel coverage reporting in CI workflow

5b4af93

Merge branch 'master' into feature/spk-baryons

66a4e9a

Jaime Salcido and others added 5 commits May 14, 2026 17:54

fix: update mathematical notation for clarity in baryons documentation

60da057

chore: remove specific Python command settings from esbonio configura…

c962888

…tion

chore: remove VSCode settings and add .vscode/ to .gitignore

55f79fd

docs: enhance parameter descriptions for BaryonsSPK class in spk.py a…

af187d5

…nd documentation

docs: update mathematical notation for clarity in BaryonsSPK class do…

ebfc187

…cumentation

jemme07 commented May 14, 2026

View reviewed changes

Conversation

jemme07 commented May 9, 2026

Overview

What is implemented

1. SP(k) wrapper interface

2. CCL integration behavior

3. Performance behavior for iterative workflows

4. Documentation and packaging updates

5. Validation tests and benchmarks

Validation run

Uh oh!

damonge left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

jemme07 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

damonge commented May 14, 2026

Uh oh!

coveralls commented May 14, 2026

Coverage Report for CI Build 25822586014

Coverage decreased (-0.3%) to 97.205%

Details

Uncovered Changes

Coverage Regressions

Coverage Stats

💛 - Coveralls

Uh oh!

damonge commented May 14, 2026

Uh oh!

damonge commented May 14, 2026

Uh oh!

jemme07 commented May 14, 2026

Uh oh!

jemme07 left a comment

Choose a reason for hiding this comment

Uh oh!

jemme07 commented May 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

jemme07 commented May 15, 2026 •

edited

Loading