Question

A data scientist is training a scikit-learn model on a single-node cluster. Their workflow first applies a StandardScaler to the full training set, then runs GridSearchCV with 5-fold cross-validation on the scaled data. A reviewer flags that this approach can produce overly optimistic validation scores. What is the correct way to restructure the code so that scaling is properly incorporated into the cross-validation process?

Accepted Answer

Combine the StandardScaler and estimator into a single sklearn Pipeline object and pass that Pipeline to GridSearchCV.

Answer

Fit the StandardScaler on the full dataset before splitting, but re-run GridSearchCV with more folds to average out any leakage.

Answer

Scale the data manually inside each fold using a for loop, then call GridSearchCV on the already-scaled folds.

Answer

Apply MinMaxScaler instead of StandardScaler, since MinMaxScaler does not leak information across folds.