```

score = cross_val_score(pipe, transformer.fit_transform(X_train), y_train, cv=5, n_jobs=-1, error_score=0.99)

```

It seems that if `transformer` includes some steps about standardization or imputation then `transformer.fit_transform(X_train)` BEFORE cross validation will have risks of leakage, right?

I guess the better way is to include `transformer` in pipe.