polars-learn

Native Polars feature engineering, schema contracts, and row-level value checks inside scikit-learn pipelines. For Python users who want the same preparation path during training and prediction: compose ordered expressions, validate declared or learned schemas, and select model columns. Supports eager and lazy Polars frames and optional pandas conversion.

Installation

Requires Python 3.11+, Polars 1.20+, scikit-learn 1.6+, NumPy 1.24+, and cloudpickle 3+. Dependencies are installed automatically.

pip install polars-learn
# Optional pandas input/output support:
pip install 'polars-learn[pandas]'

Quick start

import polars as pl

from polars_learn import FeatureEngineer, FeatureSpec

orders = pl.DataFrame({"price": [10.0, 20.0], "quantity": [2, 3]})
features = FeatureEngineer(
    FeatureSpec(pl.col("price") * pl.col("quantity"), alias="total")
)
prepared = features.fit_transform(orders)
assert prepared["total"].to_list() == [20.0, 60.0]

Expressions replay on each batch; they do not freeze learned statistics. Schema checks cover structure and dtypes, not business rules or distribution drift. Lazy queries must be collected before an estimator that requires eager input.

Read the documentation, installation and first pipeline, user guide, examples, API reference, and development guide. Licensed under MIT.