Quantcast Least squares Ridge Regression - how to do with solver that takes matrix input? Text - Physics Forums Library

PDA

View Full Version : Least squares Ridge Regression - how to do with solver that takes matrix input?


SirTristan
Jun30-08, 11:55 PM
How does one perform least squares regression with Tikhonov Regularization (http://en.wikipedia.org/wiki/Tikhonov_regularization) when using a solver that takes matrix input?

To do the regularization, you add (Gamma*x)^2 to the term that you are minimizing. But what if you are using a solver where you can't directly add this to the residual sum of squares? In my case LAPACK xgels routines.

How can one transform the matrix to incorporate the regularization term?

I read the wiki article, however I took linear algebra in college almost a decade ago and am very rusty, so I am not able to figure this out myself :smile: