[R] lda

[Pieter Vermeesch]

Dear Martin and Uwe,

I did indeed have a few -Inf values in my data frame. Few enough that
I didn't notice them when I inspected my data.

Thanks a lot for helping me better understand the MASS *package* :-)

Pieter

On 10/17/06, Martin Maechler <maechler at stat.math.ethz.ch> wrote:
> >>>>> "Pieter" == Pieter Vermeesch <pieter.vermeesch at erdw.ethz.ch>
> >>>>>     on Mon, 16 Oct 2006 19:15:59 +0200 writes:
>
>     Pieter> I'm trying to do a linear discriminant analysis on a
>     Pieter> dataset of three classes ("Affinities"), using the
>     Pieter> MASS library:
>                  ^^^^^^^
> No, no!    MASS *package*   (please!)
>
>     >> data.frame2 <- na.omit(data.frame1)
>     >>
>     >> data.ld = lda(AFFINITY ~ ., data.frame2, prior = c(1,1,1)/3)
>
>     Pieter> Error in var(x - group.means[g, ]) : missing observations in cov/cor
>
>     Pieter> What does this error message mean and how can I get rid of it?
>
> You have (+ or -) 'Inf' data values which na.omit() does not
> omit and  'x - group.means[g, ]' contains 'Inf - Inf' which is NaN.
>
> Ideally, MASS:::lda.default() would check for such a case and
> give a more user-friendly error message.
>
>     Pieter> Thanks!
>
> you're welcome.
> Martin Maechler, ETH Zurich
>


-- 
Pieter Vermeesch
ETH Zürich, Isotope Geology and Mineral Resources
Clausiusstrasse 25, NW C 85, CH-8092 Zurich, Switzerland
email: pvermees at erdw.ethz.ch, tel: +41 44 632 4643