Dear all,
an update to the "adalasso" contributed function package by myself has
been released recently, version 0.2. It implements the adaptive Lasso.
There were two reasons: First there was a bug where it was actually
impossible to choose plain OLS for the first weighting stage. (The
default there is a ridge estimator.) Secondly, there turned out to be a
little gretl bug in conjunction with the parallel computing layer MPI
(fixed in forthcoming gretl 2026a) which could be triggered by the
package. This could lead to wrong results ('nan' values for some
standard errors), now there is a workaround in place in the package, for
older gretl versions up to the current 2025c.
As a very small new feature, you can now directly get the full
coefficient vector for the "best" model (optimal tuning parameter),
including those elements that were set to zero. (Usually you just get
the subset of the non-zero coefficients, or the full matrix for all models.)
Don't hesitate to report any bugs that you come across (in this or other
packages)!
As usual, the new package version is available for download from within
the gretl program.
cheers
sven
_______________________________________________
Gretl-users mailing list -- [email protected]
To unsubscribe send an email to [email protected]
Website:
https://gretlml.univpm.it/postorius/lists/gretl-users.gretlml.univpm.it/
