let's see if I can clarify. The algo in LassoLars turns a out to be slight modification of Lars (aka LAR, stepwise regression) that actually solves the lasso problem regularized by alpha. See ESL While AFAIK the stepwise is used as an OMP ie where we would only select the number of coefs, Lasso is naturally parametrized by alpha. That's the reason why this "alpha" shows up in Lars.
There is to me no real reason why we might want to use LarsCV vs LassoLarsCV with alpha except as in Lars no feature can be dropped along the path it can be a bit faster (no cholesky downdate) ... hope this helps Alex PS : the chapter on ESL on this is pretty good. On Mon, Aug 20, 2012 at 9:30 PM, Andreas Mueller <[email protected]> wrote: > On 08/20/2012 07:31 PM, Mathieu Blondel wrote: >> With L1-regularization, for a given value of the regularization >> parameter alpha, you get a certain number of non-zero coefficients. >> Lars works by automatically computing the decrease in alpha that you >> need to increment the number of non-zero coefficient by 1. That's why >> lars_path returns the alphas, one for each feature added. > Ok, makes sense. Thanks for the explanation. > But now I don't understand what the difference between LarsCV and > LassoLarsCV is. > > Sorry don't have my ESL with me :-/ > > We should add that to the docs. There is no real explanation of LarsCV > there, right? Probably it is at lars_path? > Both lars_path and LarsCV (and the other XCV) should be added to the > narrative in the linear model section. > > Would anyone with ESL close by and / or a good knowledge of these models > volunteer? > > > ------------------------------------------------------------------------------ > Live Security Virtual Conference > Exclusive live event will cover all the ways today's security and > threat landscape has changed and how IT managers can respond. Discussions > will include endpoint security, mobile security and the latest in malware > threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/ > _______________________________________________ > Scikit-learn-general mailing list > [email protected] > https://lists.sourceforge.net/lists/listinfo/scikit-learn-general ------------------------------------------------------------------------------ Live Security Virtual Conference Exclusive live event will cover all the ways today's security and threat landscape has changed and how IT managers can respond. Discussions will include endpoint security, mobile security and the latest in malware threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/ _______________________________________________ Scikit-learn-general mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
