Machine Learning

Pipelines

The easiest way to avoid leakage and production bugs.

Why pipelines?

Same transforms in training and inference.
Less leakage (fit transforms only on training folds).
One object to save + deploy.

Minimal pipeline

from sklearn.pipeline import Pipeline
from sklearn.impute import SimpleImputer
from sklearn.preprocessing import StandardScaler
from sklearn.linear_model import LogisticRegression

pipe = Pipeline([
  ("impute", SimpleImputer(strategy="median")),
  ("scale", StandardScaler()),
  ("model", LogisticRegression(max_iter=200))
])