Bug alert in rqPen!
I have recently been working with the rqPen package in R In order to do some penalized median regression (\(L_1\) regression with a group lasso penalty) and have found a bug in the group lasso code.
As it is right now, if someone is trying to fit a quantile regression using the SCAD or MCP penalties, the function is not passing the \(\gamma\) tuning parameter (variable a in the code) to the solver.
I’ve filed an issue with the maintainer. Just wanted to flag this in case anyone is working with the rqPen package currently! Will update when/if I hear back.
Update!: I also noticed a bug when using the rq.group.fit function and trying to do a lasso penalty: it doesn’t appear to be actually doing a group penalty, just a regular lasso. I have filed another bug request.
Update (4/9/2020): I heard back from the maintainer and he has submitted an update to CRAN regarding the error for the SCAD/MCP penalties. For the group lasso, he says this is by design(!) that it doesn’t offer a group penalty, and is noted in the package help files. I misunderstood this to mean it was using some fancy linear programming method with the \(L_1\) penalty to enforce group sparsity; this belief was incorrect!
