Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
From | Kun Li <ecolikun@gmail.com> |
To | statalist@hsphsun2.harvard.edu |
Subject | Re: st: RE: One question about LARS package |
Date | Tue, 22 Jan 2013 22:23:38 +0100 |
Dear Peter, Thank you for your response. I think your suggestion would be a feasible alternative. However, such restriction would be similar to directly restricting the number of regressors at RHS, so it lacks the flexibility to have a smaller value on the tuning parameters (as I may still have different number of regressors for different data for small tuning parameter, instead of a fixed number of regressors). A further question would be, can I have a method to compare step 3, 4, and 5? Does it make sense if I compute the norm for all the coefficients at each step and compare with the parameter I set to select from these three steps? Thank you very much for your help. Best Kun 2013/1/22 Lachenbruch, Peter <Peter.Lachenbruch@oregonstate.edu>: > can you just use the model at step 3? or step 4? You'll need to rerun the regress command with those three variables. > > Peter A. Lachenbruch, > Professor (retired) > ________________________________________ > From: owner-statalist@hsphsun2.harvard.edu [owner-statalist@hsphsun2.harvard.edu] on behalf of Kun Li [ecolikun@gmail.com] > Sent: Tuesday, January 22, 2013 12:25 AM > To: statalist@hsphsun2.harvard.edu > Subject: st: One question about LARS package > > Dear Stata Users > > I have one question using LARS package in stata. I have one particular > application that tries to use lars for model selection. Basically I > have one y and x1, x2, ....up to x12. The default lars command gives > me the results as follows > > Cp, R-squared and Actions along the sequence of models > > +------------------------------------------+ > | Step | Cp | R-square | Action | > |------+-------------+----------+----------| > | 1 | 276.7940 | 0.0000 | | > | 2 | 249.3110 | 0.0292 | +x6 | > | 3 | 95.0027 | 0.1840 | +x5 | > | 4 | 13.1299 | 0.2670 | +x7 | > | 5 | 13.2321 | 0.2689 | +x8 | > | 6 | 14.7937 | 0.2694 | +x9 | > | 7 | 15.4561 | 0.2707 | +x11 | > | 8 | 14.4944 | 0.2736 | +x4 | > | 9 | 13.2267 | 0.2769 | +x3 | > | 10 | 13.6023 | 0.2785 | +x1 | > | 11 | 13.6600 | 0.2804 | +x2 | > | 12 | 12.2079 *| 0.2838 | +x10 | > | 13 | 13.0000 | 0.2850 | +x12 | > +------------------------------------------+ > > So this gives a model with 11 regressors on the RHS. However, this > number (11) is still too big for the application. Ideally it would be > around 3-5. And my question is: > > Is there a way for me to reduce the regressors on the RHS to be 3-5? > Does this have anything to do with the tuning parameter? And if so, > does LARS package allow user to adjust tuning parameters? > > Thanks for your help! > > Best > > Kun > * > * For searches and help try: > * http://www.stata.com/help.cgi?search > * http://www.stata.com/support/faqs/resources/statalist-faq/ > * http://www.ats.ucla.edu/stat/stata/ > * > * For searches and help try: > * http://www.stata.com/help.cgi?search > * http://www.stata.com/support/faqs/resources/statalist-faq/ > * http://www.ats.ucla.edu/stat/stata/ * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/faqs/resources/statalist-faq/ * http://www.ats.ucla.edu/stat/stata/