Reproducing glmnet results runs into issues with indicator variables and ultrasparse lambdas about cvxr HOT 3 CLOSED

Thank you! I thought about it after I posted. Now I've re-run setting standardize=FALSE in glmnet and got perfect agreement. However, in your tutorial examples page these issues are not mentioned. I simply followed the example. Thank you so much, Georgina Den fredag 22 december 2023 kl. 22:39:35 CET, Balasubramanian Narasimhan ***@***.***> skrev: You have to be careful when comparing with glmnet results, because glmnet standardizes the design matrix but returns the coefficients on the original scale. See the standardization documentation of the glmnet function. So if you standardize the design matrix and pass to CVXR and multiply the returned coefficients by the standard deviations, it will match closely. The intercept will also have to be appropriately transformed by means of an easy formula that you can derive. Further note that glmnet uses active set methods, so coefficients can be exact zeros. That may not be the case with CVXR, unless one uses an active set solver with CVXR. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***>