The same evaluation discipline applies whether the estimator is simple or penalized. These tasks mix standalone scaling with ridge and lasso pipelines on fresh fictional duration records. Read each requested alpha and preserve the original training and test split. A pipeline fits the scaler first and passes the transformed training matrix to the estimator.
The holdout must use the training scale. Its own mean will usually differ from zero after transformation, which is expected. One task adds a constant input so you can check that the transformation remains finite. The other tasks vary penalty type and strength. There are no new estimators here; the challenge is keeping the preparation, fitting and prediction sequence consistent while the data and requested settings change.