[R-sig-ME] Glmmlasso - Fisher matrix not invertible

[Samuel Muli]

Hi,

I am fitting a glmmLasso model for feature selection before performing a 
traditional LMM for inference. My problem is that while running my 
model, it encounters a fatal error, "fisher matrix not invertible" just 
before returning the results (see Trace back results leading to the error).

I have tried to solve this problem by reducing the number of features, 
and all worked perfectly if my features are below ~110. I have 54 unique 
measurements at two time points (and since this is publicly available 
dataset,  I attach it here for your quick review, along with my code).   
However, since I would like to compare Glmmlasso with other methods, I 
do not want to do this prescreening (using another method).

Based on my research, another author observed similar error but 
somewhat, managed to sort it when regression could not be completed, see 
https://biodatamining.biomedcentral.com/articles/10.1186/s13040-018-0173-9 
(p. 7). Their R code is available 
https://github.com/ghedin-lab/LassoGLMMforMicrobiomes

I would really appreciate insights into this problem.


Best regards,





-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: my_code.txt
URL: <https://stat.ethz.ch/pipermail/r-sig-mixed-models/attachments/20220716/9e2fdadc/attachment-0001.txt>