Re:An Introduction to Gradient Descent and Linear Regression
by babs474 on Thursday June 26, 2014 @01:51PM
(#7950778)
You need it for l1 and l2 regularization which is popular these days don't you?
[ Parent
| on HN ]
Re:An Introduction to Gradient Descent and Linear Regression
by throwaway283719 on Thursday June 26, 2014 @01:53PM
(#7950790)
No. L2 regularization (ridge regression) also has an analytical solution, and L1 regularization (lasso) can be more efficiently solved using the LARS (least-angle regression) algorithm.
Gradient descent would work for ridge regression (it would just be slow). It wouldn't even work for lasso, because the penalty function isn't differentiable at the origin.