Machine Learning · Chapter 35 of 40

Pipelines

A PIPELINE chains preprocessing and modeling into one object. Prevents data leakage and simplifies deployment.

Use `ColumnTransformer` to apply different steps to different columns.

Example 1 (python)

from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.linear_model import LogisticRegression
p = Pipeline([('sc', StandardScaler()), ('lr', LogisticRegression())])
p.fit(X_train, y_train)

Scale then model — one call.

Example 2 (python)

# ColumnTransformer for mixed feature types
from sklearn.compose import ColumnTransformer

Different preprocessing per column set.

Key points

Chain preprocessing + model.
Prevents data leakage in CV.
One object to serialize.
Use ColumnTransformer for mixed types.

💡 Note: Grid searching hyperparameters of any step is possible via `pipeline.set_params(step__param=value)`.

Pipelines

Key points

📝 Quick Quiz