glmnet
Lasso and Elastic-Net Regularized Generalized Linear Models
Extremely efficient procedures for fitting the entire lasso or elastic-net regularization path for linear regression, logistic and multinomial regression models, Poisson regression, Cox model, multiple-response Gaussian, and the grouped multinomial regression; see <doi:10.18637/jss.v033.i01> and <doi:10.18637/jss.v039.i05>. There are two new and important additions. The family argument can be a GLM family object, which opens the door to any programmed family (<doi:10.18637/jss.v106.i01>). This comes with a modest computational cost, so when the built-in families suffice, they should be used instead. The other novelty is the relax option, which refits each of the active sets in the path unpenalized. The algorithm uses cyclical coordinate descent in a path-wise fashion, as described in the papers cited.
README
# Changes in Mortran and R
## Tracing progress fixes in Mortran and R
1. Changed the function definition
```
subroutine get_int_parms(sml,eps,big,mnlam,rsqmax,pmin,exmx);
```
to
```
subroutine get_int_parms(sml,eps,big,mnlam,rsqmax,pmin,exmx,itrace);
```
taking on itrace as a switch for tracing progress. Added an `entry
chg_itrace` to change `itrace`.
2. Changed all calls to `get_int_parms` to include extra parameter.
3. Corresponding changes to `glmnet.control.R` to include `itrace` as a
parameter.
4. Fixes in `relax.glmnet.R` for displaying the progress by
temporarily turning off and on tracing, if it was on.
## Unused variable fixes in Mortran.
1. The subroutines `spstandard`, `spstandard1`, `multspstandard1` had `jerr` as an unused
argument.
2. The subroutines `coxnet1`, `fishnet1`, `spfishnet1` had `isd` as an
unused argument.
3. The subroutine `risk` had `ni` as an unusued argument.
In all these cases, I merely added a _no-op_ statement to the
function/subroutine (e.g. `jerr = jerr * 1` or `ni = ni * 1`) to stop
compiler warnings. More advanced compilers may detect these
shenanigans. We can address that later if it comes up.
## Some extraneous tabs in Mortran.
1. The subroutine `multstandard1` has an extra tabl between the
arguments `ju` and `xm` that I removed. This is most likely because
Emacs replaces 8 spaces with tabs.
## Uninitialized variable in Mortran
1. The `gfortran` compiler warns that the variable `alf` is
uninitialized in `elnet1`. The logic is correct, but the compiler
is unable to detect it. So I initialized `alf=1` and combined it
with a previous fix for `alm=0.0` in `elnet1`. I am reasonably
certain this will not affect any logic.
## Generating Fortran from Mortran
Note that it is Mortran processing that generates unusued labels. So
each time you run Mortran to generate the Fortran, you will have to
deal with the problem of unused labels. No worries, however, if you
follow this procedure.
1. Install the latest [`SUtools`](https://github.com/bnaras/SUtools)
package, after deleting your old one, if you had it.
2. Execute the following, assuming `GLMNET_DIR` is your `glmnet`
package directory.
```{r, eval = FALSE}
library(SUtools)
GLMNET_DIR <- "~/GITHUB/glmnet"
mortran <- readLines(file.path(GLMNET_DIR, "inst", "mortran", "glmnet5dpclean.m"))
result <- process_mortran(input_mortran_file = "glmnet5dpclean.m",
pkg_name = "glmnet",
control = sutools_control(fix_allocate = FALSE))
```
The fortran file can now be written out to a file. For example, you
may choose to overwrite the old fortran with the new fortran.
```{r, eval = FALSE}
writeLines(fixed_fortran, file.path(GLMNET_DIR, "src", "glmnet5dpclean.f"))
```
## Maxit bug fixed
Mortran would branch to :again: when convergence was not reached, and
not check if maxit had been reached. On the new mac M1 this could get
into a state with a degenerate example where it never
terminated. Added a maxit check immediately after again throughout.
Versions across snapshots
| Version | Repository | File | Size |
|---|---|---|---|
4.1-10 |
2026-04-09 windows/windows R-4.5 | glmnet_4.1-10.zip |
2.3 MiB |