sm00thix / ikpls / 28707813976
94%

DEFAULT BRANCH: main
Ran
Jobs 15
Files 13
Run time 1min
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst
coverage: 94.323% (+0.8%) from 93.529%
28707813976

push

github

web-flow 
IKPLS 6.0.0: scikit-learn-conformant PLS wrapper, backend-named modules, ddof=0 default (#47)

* IKPLS 6.0.0: scikit-learn-conformant PLS wrapper, backend-named modules, ddof=0 default

This is a breaking major release. It adds a fully scikit-learn-conformant PLS,
reorganizes the package into a consistent backend-named layout, changes the
default scaling degrees-of-freedom, exposes PLS Y-weights, makes submodule
imports lazy, and hardens the JAX implementations against degenerate fits.

New module layout (breaking):
- Backend-named public modules: ikpls.numpy.PLS, ikpls.jax.PLS(algorithm=1|2),
  ikpls.sklearn.PLS, and ikpls.fast_cross_validation.numpy / .jax. The
  algorithm-specific JAX classes and the shared NumPy inner loop moved to a
  private ikpls._impl/ subpackage (via git mv; history preserved).
- ikpls.jax.PLS is a dispatcher: it subclasses the shared JAX base (so the class
  carries and documents the full public API) and its __new__ returns the
  algorithm-specific implementation registered as a virtual subclass. Hence
  isinstance(model, ikpls.jax.PLS) / issubclass hold, model.algorithm is 1 or 2,
  and model.stateless_fit / jax.vmap run directly on the implementation with no
  wrapping overhead (verified: warm vmap speedup unchanged at ~16.4x, results
  bit-identical).

New ikpls.sklearn.PLS -- scikit-learn PLSRegression conformance:
- A thin wrapper around ikpls.numpy.PLS mirroring
  sklearn.cross_decomposition.PLSRegression: a conformant regressor AND
  transformer that passes check_estimator with no expected failures on
  scikit-learn 1.6 through 1.9. Exposes the PLSRegression fitted attributes
  (x_weights_, x_loadings_, y_weights_, y_loadings_, x_rotations_, y_rotations_,
  coef_, intercept_, n_features_in_), supports transform/inverse_transform of
  both X and Y, implements score(X, y, sample_weight=None), and preserves the
  (A, N, M) all-components prediction via predict_all_components.
- y_weights_ == y_loadings_, and coef_/interc... (continued)

1213 of 1286 relevant lines covered (94.32%)

8.52 hits per line

Coverage Regressions

Lines Coverage ∆ File
3
80.0
 8.57% fast_cross_validation/__init__.py
3
81.25
 6.25% __init__.py
Jobs
ID Job ID Ran Files Coverage
1 windows-latest-python-3.13-jax - 28707813976.1 13
93.39
 GitHub Action Run
2 ubuntu-latest-python-3.11-numpy-only - 28707813976.2 13
24.73
 GitHub Action Run
3 ubuntu-latest-python-3.12-jax - 28707813976.3 13
93.39
 GitHub Action Run
4 windows-latest-python-3.13-numpy-only - 28707813976.4 13
24.81
 GitHub Action Run
5 ubuntu-latest-python-3.14-numpy-only - 28707813976.5 13
24.81
 GitHub Action Run
6 windows-latest-python-3.11-jax - 28707813976.6 13
93.39
 GitHub Action Run
7 ubuntu-latest-python-3.13-jax - 28707813976.7 13
93.39
 GitHub Action Run
8 windows-latest-python-3.12-numpy-only - 28707813976.8 13
24.73
 GitHub Action Run
9 ubuntu-latest-python-3.13-numpy-only - 28707813976.9 13
24.81
 GitHub Action Run
10 windows-latest-python-3.11-numpy-only - 28707813976.10 13
24.73
 GitHub Action Run
11 ubuntu-latest-python-3.12-numpy-only - 28707813976.11 13
24.73
 GitHub Action Run
12 windows-latest-python-3.12-jax - 28707813976.12 13
93.39
 GitHub Action Run
13 ubuntu-latest-python-3.14-jax - 28707813976.13 13
93.39
 GitHub Action Run
14 windows-latest-python-3.14-numpy-only - 28707813976.14 13
24.81
 GitHub Action Run
15 windows-latest-python-3.14-jax - 28707813976.15 13
93.39
 GitHub Action Run
Source Files on build 28707813976