grLasso does not work with cv.grpregOverlap for a smaller dataset

[computingmachine]

Hi:
I tried to use cv.grpregOverlap with grLasso,grMCP and grSCAD for same dataset(6000 observations and 500 variables ) and group to fit the logistic regression. The three penality all work well.

But when I subset the dataset with less variables and redefine the group, grLasso just returns NA for all the coefficients. grMCP and grSCAD still works well.

In addition, I also tried the function grpregOverlap for both the original set and the subset, grLasso works for both conditions.

Why grLasso does not work when I just reduce the number of variables and using cross-validation?

Could you please check this? Thanks!

[YaohuiZeng]

Hi @computingmachine , sorry for the so delayed response. Will it be possible that you send me a MWE so that I can take a look? Thanks!

[CharlieCheckpt]

Hi @YaohuiZeng, I encountered the same problem recently.
It seems to be due to the fact that my dataframe used for running cv.grpsurvOverlap has column with constant value (column of zeros).
It would be great if you could fix this !