R help - Dec 2003 - Fast logistic regression

Is there any faster implementation of logistic regression
than glm function in base package in R?

I am working on simulation study and discovered glm fitting part is the
bottle neck of my program.

-Masahiko

On Thu, 25 Dec 2003, Masahiko AIDA wrote:
> Is there any faster implementation of logistic regression
> than glm function in base package in R?
> 
> I am working on simulation study and discovered glm fitting part is the
> bottle neck of my program.
Have you tried profiling (see Writing R Extensions)?  It may not be the
actual fitting but, say, the extraction of the model matrix which is slow.  
You could call glm.fit() directly if it produces all you need.  In a
well-behaved problem you are not going to get anything much faster than
glm.fit without going entirely to compiled code (which is an option for
you, of course).

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595

On Thu, 25 Dec 2003 11:53:40 +0900
Masahiko AIDA <maida at umich.edu> wrote:
> Is there any faster implementation of logistic regression
> than glm function in base package in R?
> 
> I am working on simulation study and discovered glm fitting part is the
> bottle neck of my program.
> 
> -Masahiko
For this purpose (e.g., bootstrapping model fits using the bootcov
function) I use the lrm.fit function in the Design package.  It assumes
the model matrix is already created and that there are no NAs.

Frank

---
Frank E Harrell Jr   Professor and Chair           School of Medicine
                     Department of Biostatistics   Vanderbilt University