Comments on: Regularized Least Squares

By: Mohsen

Mohsen — Tue, 02 Oct 2012 22:30:59 +0000

Was a great post. I am struggling with kernel ridge for a while. I have a question regarding the regularizer. As far as I have read there two method for regularization of kernel matrix. One is , adding a small value to the diagonal of kernel matrix.and the second one is using truncated eigendecomposition which means just to put out the small eigenvalues .But I kind of stuck in the second case.I actually can’t understand why it’s a case.I know it kind of related to matrix inversion.Do you have any idea about that?

By: Stratis

Stratis — Mon, 31 Oct 2011 11:15:00 +0000

Hi!

Very interesting post. I have one question though, is there any error analysis w.r.t. ridge regression? What happens when for example the parameter vector w gets less and less sparser.

By: Mathieu

Mathieu — Sun, 13 Feb 2011 16:12:23 +0000

That would be nice! I have much to learn about LARS.

By: Fabian Pedregosa

Fabian Pedregosa — Sun, 13 Feb 2011 12:05:08 +0000

This is awesome, it really motivates me to write something similar about the Least Angle regression implementation in the scikit.

Cheers,

Fabian.

By: Mathieu

Mathieu — Sat, 12 Feb 2011 19:11:00 +0000

Sure, will do that next week!

By: Gael Varoquaux

Gael Varoquaux — Sat, 12 Feb 2011 10:19:40 +0000

Nice post (as usual).

Any chance that you can contribute a very condensed version of it to the scikit’s documentation? I am thinking in particular that the GCV should be mentioned in the documentation.