FeaturePassthrough workflows¶

About¶

An explicit unchanged-feature step selects named columns while preserving their Polars values, names, dtypes, nulls, and row order. It communicates the selection contract more clearly than an anonymous passthrough string.

Why use it?¶

Use it for features that need no transformation, particularly a ColumnTransformer branch alongside transformed features. In a sequential pipeline it returns only the named columns, so the next step receives that subset.

Examples¶

Input¶

Code¶

import polars as pl

from polars_learn import FeaturePassthrough

raw = pl.DataFrame({"id": [7, None], "flag": [True, False], "unused": [9, 8]})
keep = FeaturePassthrough(["flag", "id"])
prepared = keep.fit_transform(raw)
assert prepared.equals(raw.select("flag", "id"))
assert prepared.to_dict(as_series=False) == {"flag": [True, False], "id": [7, None]}
assert dict(prepared.schema) == {"flag": pl.Boolean, "id": pl.Int64}

Output¶

The assertions in the example verify the output against the implementation. Python examples are executed during documentation validation.

Practical notes¶

There is no rename or cast option. Input order and dtypes may change without rejection; add schema validation when those changes must fail. See parallel branch examples for combining unchanged and transformed features.