Machine Learning 18: Regularization

We present regularization as a technique for combatting overfitting. As a concrete example, we introduce weight decay and show how to modify the linear regression learning algorithm when weight decay is added to the loss function, resulting the ridge regression algorithm. We also show in an experiment how just a little regularization may greatly improve the quality of a learned model in the presence of noise.

[kylmaz]

Hello. Which method should I follow to get the best lambda value? For example I have a small dataset and I want to manually (by hand) get the best lambda value. How do I do this?