quantile_lasso or gurobi seems sensitive to floating number precision
jingjtang opened this issue · comments
Jingjing Tang commented
We ran into the error
Error in .local(x, i, j, ..., value) :
not-yet-implemented 'Matrix[<-' method
when using quantile_lasso
using test.rds.
This is the script that I used test.R
We will see the error mentioned above if run the training with quantile_lasso
. If we explore further into the error, we will know that gurobi
tells us that this problem is INFEASIBLE
.
However, we won't run into the same error if we do either of the thing listed below:
- replace
tau=0.9
to tau=0.900001
- replace
gurobi
withglpk
- change the precision of the floating number by saving and re-loading the data using
save.csv
andread.csv
This seems weird.