On 12.02.2014 15:55, ben1983 wrote:
Hi All,
I've been trying to write my own code for LDA (linear discrim) so I
can modify it to be weighted LDA since some of my groups are outliers.
However, the code I write for standard LDA gives me slightly different
results to those from R (slightly different LDAs.....and not just scalar
differences.
Does R simply invert the pooled within groups scatter matrix e.g. solve(Sw)
and multiply it by the between groups scatter Sb, then take the eigen
vectors?
This is the approach that I use, yet get slightly different
results..........does R by default use a slightly more robust inverse or
estimate of Sw??
Neither nor, the actual algorithm is more tricky than that.
Swetlana Herbrandt, one of my former master students, wrote her master
thesis about the lda algorithm used in MASS. The thesis is written in
German, but the formulas can be read anyway. If you are interested,
please ask Swetlana (CCed) for a PDF version.
Best,
Uwe Ligges
I've looked at the source code, and I find it very hard to follow.
If anyone has any ideas that would be great!!
Thanks,
Ben
--
View this message in context:
http://r.789695.n4.nabble.com/Actual-code-for-LDA-tp4685194.html
Sent from the R help mailing list archive at Nabble.com.
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
