Selecting model complexity on test data: $($select all that apply$)$

$-$Allows you to avoid issues of overfitting to training data

$-$Should never be done

$-$Provides an overly optimistic assessment of performance of the resulting model

$-$Is computationally inefficient