Scikit-Learn Pipeline

Also known as · Pipeline

A single scikit-learn object that chains preprocessing steps (such as a scaler) and a model, so the preprocessing is fitted only on training data and Data Leakage is prevented.

When to use

Use one whenever you cross-validate with scaling: fitting the scaler on the full dataset first leaks test information, whereas inside a pipeline it is re-fitted fresh for each fold. The lecture's full workflow is scaler, then Ridge, then cross_val_score. Also in PS 02.

See DS Lec 4.

Appears in