Hello,
I am using the a-score to determine the optimal number of PCs to retain in a
discriminant function of principal components analysis, and am trying to figure
out whether the result I am getting is pathological or not.
I am trying to find the discriminant axes between two groups. One group is much
larger than the other. The variances explained by the PCs do not fall off
rapidly--the first 58 PCs explain ~95% of the variance.
When I run optim.a.score(), I get a decaying curve without an internal optimum.
The function says retaining only a single PC is optimal, with an a-score of
~0.4. I thought this might be an artifact of default values, to I increased the
number of PCs at which to conduct simulations (n option) and increased the
number of simulations per PC point to 50 (from 10), but I still get the
boundary answer of 1 PC.
Is this valid, or might this be a pathology of diffuse PC variation
distribution and asymmetric sample sizes?
Thanks!
~John
[[alternative HTML version deleted]]
_______________________________________________
R-sig-genetics mailing list
R-sig-genetics@r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-genetics
