Black Swans


5 points

Hey there! A sophomore CS undergrad, currently going through Introduction to Statistical Learning (ISLR). Solutions to exercises :

History focuses on a top-down approach. They recently launched a ML MOOC too.

Transforming response vs Weighted least square fit in case of heteroscedasticity

In the case of Linear Regression models, some data has residual terms influenced by a predictor term (heteroscedasticity). Taking a case where the residual increases with increase in predictor value, transforming the data using a concave function like log Y might help.
But how does this impact other aspects of the model? Is there a disadvantage? Is a weighted least squares fit better in this case?

Awesome! Would love to be a part of it.

Awesome! Would love to be a part of it.

I came across this a while back, still in draft though :