Hi. I have successfully performed the discriminant analysis with the lda function, I can classify new individuals with the predict function, but I cannot figure out how the lda results translate into the classification decision. That is, I don't realize how the classification equation for new individuals is constructed from the lda output. I want to understand it but also, I need to communicate it and provide a mechanism for other colleagues to make classifications with their data. Thank you very much, Fernando -- *Dr. Fernando M. Archuby* CONICET-UNLP Tel?fono Personal: +54-221-15-6129667. farchuby at gmail.com paleobiologia at gmail.com [[alternative HTML version deleted]]
It's possible that neither of these will help, but
(1) you can look at the source code of the predict method
(MASS:::predict.lda)
(2) you can look at the source reference ("Modern Applied Statistics in
S", Venables and Ripley) to see if it gives more information (although
it might not); there's a chance that you can get the information you
need via a google books search
On 2023-10-12 10:25 a.m., Fernando Archuby wrote:> Hi.
> I have successfully performed the discriminant analysis with the lda
> function, I can classify new individuals with the predict function, but I
> cannot figure out how the lda results translate into the classification
> decision. That is, I don't realize how the classification equation for
new
> individuals is constructed from the lda output. I want to understand it but
> also, I need to communicate it and provide a mechanism for other colleagues
> to make classifications with their data.
> Thank you very much,
> Fernando
>
On 12.10.2023 16:25, Fernando Archuby wrote:> Hi. > I have successfully performed the discriminant analysis with the lda > function, I can classify new individuals with the predict function, but I > cannot figure out how the lda results translate into the classification > decision. That is, I don't realize how the classification equation for new > individuals is constructed from the lda output. I want to understand it but > also, I need to communicate it and provide a mechanism for other colleagues > to make classifications with their data. > Thank you very much, > Fernando >Do you want to know the principles of the theory behind LDA? That is available in lots of textbooks. Do you want the implementation detials of MASS::lda()? That is hard. It is based (but does not follow in all details) on a paper by Nils Hjort from Norway. A former student of mine, Swetlana Herbrandt, has analysed and reverse engineered the code and wrote down the theory in a German thesis. The implementation uses some nice tricks to get numerically rather stable results that are typically not mentioned in any textbook. Do you really want to do prediction with LDA? I typically look at classificatuion performance of LDA as a reference to compare better and more modern techniques with. I think you should ask some trained local statistician for advise on both, the LDA theory and for prediction in general. Best, Uwe Ligges