R help - Aug 2003 - LDA in R: how to extract full equation, especially constant term

Hi,

Having dipped my toe into R a few times over the last year or two, in the 
last few weeks I've been using it more and more; I'm now a thorough 
convert. I've just joined the list, because although it's great, I do
have
this problem...

I'm using linear discriminant analysis for binary classification, and am 
happy with the classification performance using predict(). What I'd like to 
do now is extract the equation for this classifier, for use elsewhere (in 
Perl/Python code).

I know that I can get the means and scaling factors from the predict() 
object, but I'm having trouble computing the constant term. From reading 
Venables & Ripley and Hastie/Tibshirani/Friedman, I know the priors play 
a  role in adjusting the "cut-point" from zero (for equally sized
classes),
based on the relative sizes of the two classes. But when I try to do the 
computation, I don't get a value that agrees with that returned by
predict().

I've seen a post about this problem in the past, but it was never really 
answered by anyone who was familiar with R/S-PLUS. Can anyone help me with 
this? I guess I'm really wondering how R is computing the constant term in 
its discriminant function.

Thanks,

-Frank Gibbons

PhD, Computational Biologist,
Harvard Medical School BCMP/SGM-322, 250 Longwood Ave, Boston MA 02115, USA.
Tel: 617-432-3555       Fax: 
617-432-3557       http://llama.med.harvard.edu/~fgibbons

You have the R code: please read it.  Hint: these isn't `an equation',
but
LDA chooses the largest of several expressions, and those expressions are
in all the standard books, including V&R and in more detail in my PRNN 
book.  For numerical stability reasons the `constants' are adjusted to 
keep the largest expression finite in computer arithmetic.

On Thu, 21 Aug 2003, Frank Gibbons wrote:
> Hi,
> 
> Having dipped my toe into R a few times over the last year or two, in the 
> last few weeks I've been using it more and more; I'm now a thorough
> convert. I've just joined the list, because although it's great, I
do have
> this problem...
> 
> I'm using linear discriminant analysis for binary classification, and
am
> happy with the classification performance using predict(). What I'd
like to
> do now is extract the equation for this classifier, for use elsewhere (in 
> Perl/Python code).
> 
> I know that I can get the means and scaling factors from the predict() 
> object, but I'm having trouble computing the constant term. From
reading
> Venables & Ripley and Hastie/Tibshirani/Friedman, I know the priors
play
> a  role in adjusting the "cut-point" from zero (for equally sized
classes),
> based on the relative sizes of the two classes. But when I try to do the 
> computation, I don't get a value that agrees with that returned by
predict().
> 
> I've seen a post about this problem in the past, but it was never
really
> answered by anyone who was familiar with R/S-PLUS. Can anyone help me with 
> this? I guess I'm really wondering how R is computing the constant term
in
> its discriminant function.
> 
> Thanks,
> 
> -Frank Gibbons
> 
> PhD, Computational Biologist,
> Harvard Medical School BCMP/SGM-322, 250 Longwood Ave, Boston MA 02115,
USA.
> Tel: 617-432-3555       Fax: 
> 617-432-3557       http://llama.med.harvard.edu/~fgibbons
> 
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://www.stat.math.ethz.ch/mailman/listinfo/r-help
> 
> 
-- 
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