skpoly

skpoly provides drop-in polynomial feature generators that integrate with scikit-learn pipelines. The library focuses on smooth orthogonal bases such as Bernstein and Legendre polynomials, letting you capture non-linear structure with well-conditioned numerical behavior.

The project expands on the the blog series, beginning with the post “Are polynomial features the root of all evil?”.

Documentation

Explore the documentation.

Quick start

Create a pipeline that first rescales each input dimension and then expands it with Bernstein features before fitting a linear model:

from sklearn.pipeline import make_pipeline
from sklearn.preprocessing import MinMaxScaler
from sklearn.linear_model import Ridge
from skpoly import BernsteinFeatures

pipeline = make_pipeline(
    MinMaxScaler(),
    BernsteinFeatures(degree=8),
    Ridge(alpha=1e-2),
)

pipeline.fit(X_train, y_train)
y_pred = pipeline.predict(X_test)

The MinMaxScaler step keeps every feature inside the default [0, 1] range assumed by the polynomial bases, which in turn preserves the well-conditioned behavior of the Bernstein and Legendre transforms.

Pairwise interaction features

For multivariate inputs you can enable tensor-product features to model pairwise (and higher-order) interactions between coordinates. Setting tensor_product=True expands the basis with every combination of the univariate polynomials:

from sklearn.pipeline import make_pipeline
from sklearn.preprocessing import MinMaxScaler
from sklearn.linear_model import LogisticRegression
from skpoly import LegendreFeatures

pipeline = make_pipeline(
    MinMaxScaler(),
    LegendreFeatures(degree=5, tensor_product=True),
    LogisticRegression(),
)

Development

Clone the repository and install dependencies using uv:

git clone https://github.com/alexshtf/skpoly.git
cd skpoly
uv venv
source .venv/bin/activate
uv sync

Using uv keeps dependency resolution fast and reproducible.

Citation

• Use the CITATION.cff file. Most reference managers and services like GitHub's "Cite this repository" option can import the citation metadata directly from CITATION.cff.
• Grab the BibTeX entry. If you prefer to add the reference manually, cite the project as follows.

@software{Shtoff_skpoly_2025,
  author = {Alex Shtoff},
  title = {skpoly: Polynomial basis transformers for scikit-learn},
  url = {https://github.com/alexshtf/skpoly},
  version = {0.1.0},
  year = {2025}
}