Why regularization term is not included when calculating error

About the part advice in applying machine learning, regularization term is not included when calculating error. I believe the reason is we need to report the true estimate how well a model performs. Therefore, the regularization term is excluded. Am I right about this?

Hi @empheart ,

Yes, that is correct if you are referring to the development and test sets. The development and test sets are not used for training, so it is irrelevant to use the regularization term for calculating the loss because these sets are used for a fair estimation of the model’s performance on unseen data. However, the regularization term is used with the training error during training for calculating the regularized cost.

Regards,
Mujassim

2 Likes