[R-pkg-devel] Getting summary.glm to use Matrix::chol2inv

[Duncan Murdoch]

I think the only way to do this is to write your own summary.* method 
for your objects.  You should modify their class (e.g. to `c("glm2", 
"glm")` so that your new summary method `summary.glm2` is called instead 
of `summary.glm`.

Changing the search path wouldn't work.  Search order from within a 
package looks at locals, then imports, then the base package (where 
chol2inv would be found), then the search list.

The source for the standard summary.glm function is available, e.g. 
here: 
https://github.com/wch/r-source/blob/cfda9838f261b1c77c5ac6a5b0deed88021a2156/src/library/stats/R/glm.R#L687-L773 
.  You can copy that into your package, then modify it to work with 
Matrix.  You should list "The R Core Team" as a copyright holder on that 
file if you do this so you're not violating the R license.

Duncan Murdoch


On 2024-12-22 4:26 p.m., Mark Donoghoe wrote:
> Hi,
> 
> I am trying to implement a new method for glm2 which allows the use of
> the Matrix package to work with large data.
> 
> I have added Matrix to Suggests, and when the new method is requested,
> the following check is performed:
> 
> if (!requireNamespace("Matrix", quietly = TRUE)) {
>        stop("Package \"Matrix\" must be installed to use method =
> \"glm.fit2.Matrix\".",
>             call. = FALSE)
>      }
> 
> This works fine. However, I would like the user to be able to use
> summary.glm on the fitted object x, and it effectively calls
> chol2inv(x$qr$qr[1:p,1:p]), which fails if x$qr$qr is not a (small m)
> matrix:
> 
> Error in chol2inv(Qr$qr[p1, p1, drop = FALSE]) :
>    'a' must be a numeric matrix
> 
> I can return the matrix version of x$qr$qr, but this can be quite
> large and ideally I would like to return the Matrix version, and have
> summary.glm automatically call the Matrix version of chol2inv.
> 
> I imagine this would require changing the search path, which I
> understand is not recommended, but I am not even sure how to go about
> it, or if there is another solution that wouldn't require a custom
> version of summary.glm. I also tried attachNamespace in a similar code
> block when glm.fit2.Matrix is called, but that did not work.
> 
> Thanks in advance for any suggestions.
> 
> Mark
> 
> ______________________________________________
> R-package-devel using r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-package-devel