03·Pipeline — preprocessing and model in one object

Introduction

Pipeline in sklearn is not just convenience — it is an architectural guarantee against data leakage. It combines transformers (scaler, encoder, PCA, feature selector) with a final predictor into ONE object that behaves like a regular estimator: fit/predict/score/get_params. Thanks to this, inside cross_val_score and GridSearchCV all preprocessing is fit ONLY on fold-train. This lesson systematises: Pipeline vs make_pipeline, ColumnTransformer for mixed feature types, accessing steps via named_steps, the step__param syntax in grid search, serialisation through joblib. After it you will never write a manual "scaler → model → evaluate" loop again.