[R] LDA

[Prof Brian Ripley]

On Tue, 1 Apr 2003, array chip wrote:

> 
> I used the "lda" function in the MASS library of S-Plus (R) to do a
> linear discriminant analysis, and got the linear coefficients, say b1
> and b2 for the 2 predictors x1 and x2. I have trouble to calculate the
> discrimiant scores for each observation, I used 3 ways to try to repeat
> the scores returned by the "predict" function in S-Plus:
> 
> 1. b1*x1+b2*x2
> 2. b1*(x1-mean of x1)+b2*(x2-mean of x2)
> 3. b1* standardized x1+b2*standardized x2 (standardize: mean 0 & variance 1)
> 
> none of the above procedures can repeat the scores returned by the
> "predict" function. However, method 2 & 3 can predict the classes
> correctly if using 0 as cutoff, juts like using the "predict function".

You've sent this to R-help, but S-PLUS (sic) and R are different as is my 
lda() function in each.  MASS is a book, and it contains the details, 
as does the code.

> What should be the correct formula to compute the scores for each
> observation?
> 
> BTW, how to retrieve the linear coefficients from an lda object? I can't
> retrieve it by using @coef, @coefficients, etc.

Of course not: in R these are not S4 classes.  You could read the help 
page to see what lda() actually computes.

-- 
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