DSA Studio

Regularization Checkpoint

Overfitting, L1, L2 regularization.

Get every question correct to pass this checkpoint. You can review feedback, revise your answers, and submit again.

1. Loss includes lambda*sum(w^2), with lambda=0.1 and w=3. What is the penalty gradient at w?

Select one answer.

2. Increasing regularization raises both training and validation error substantially. What is a plausible diagnosis?

Select one answer.

3. Training error keeps falling but validation error starts rising. Which saved model does validation-based early stopping prefer?

Select one answer.

4. For weights [2,-3], what is the unscaled L1 penalty?
5. A standardizer is fitted to all data before the train/test split. What should change?

Select one answer.

Join Discord