There’s one wide-spread statement in statistics and machine learning literature that increased flexibility comes with decreased robustness in the test set. However, is it always true?

This issue is related to Sklearn, one solution is to state a eligible Solver .