We used to work with a complicated set of conditions that define step size increases for the shrinkage parameter lambda in lasso. In the paper we initially submitted to JOP, these conditions were
Step size increase
Lambda values
Number of iterations w/o improvement
+ 0.1
0 <= lambda < 1
NA
+ 0.3
1 <= lambda < 10
NA
+ 1
10 <= lambda < 10,000
NA
+ 10
100 <= lambda < 10,000
> 45
together with the general stoping rule:
maximum number of iterations w/o improvement = 60.
In the package, I now simplified this to
Step size increase
Lambda values
+ 0.1
0 <= lambda < 1
+ 0.3
1 <= lambda < 10
+ 1
10 <= lambda < 10,000
+ 10
100 <= lambda < 10,000
also with the general stoping rule:
maximum number of iterations w/o improvement = 60.
In case this decreases performance of lasso, we might need to revert to the more complicated set of conditions.
We used to work with a complicated set of conditions that define step size increases for the shrinkage parameter lambda in lasso. In the paper we initially submitted to JOP, these conditions were
together with the general stoping rule: maximum number of iterations w/o improvement = 60.
In the package, I now simplified this to
also with the general stoping rule: maximum number of iterations w/o improvement = 60.
In case this decreases performance of lasso, we might need to revert to the more complicated set of conditions.