So, in this session:
- We discussed the need for regularisation, which helps models perform well with unseen data while identifying necessary underlying patterns in it. We did this by adding a penalty term to the cost function used by OLS.
- We discussed two methods, Ridge and Lasso regression, which both allow some bias to get a significant decrease in variance, thereby pushing the model coefficients towards 0.
- You learnt that in Lasso, some of these coefficients become 0, thus resulting in model selection and, hence, easier interpretation, particularly when the number of coefficients is very large.