Michael Friendly wrote:
Last week I asked about data ellipses with rgl:::ellipse3d() with lines
showing the principal axes.
(The goal is a visual demonstration of PCA as a rotation of variable
space to component space.)
I was trying, unsuccessfully, to use princomp() to generate the PCA axes
and plot them using
segments3d:
PC <- princomp(trees)
sdev <- PC$sdev # component standard deviations
sd <- sqrt(diag(cov)) # variable standard deviations
# vectors in variable space of principal components
vec <- matrix(mu,3,3, byrow=TRUE) + diag(sd) %*% PC$loadings
for (j in 1:3) {
mat <- rbind(mu, vec[j,])
segments3d(mat, col="red")
}
However, it occurred to me that these axes are just the orthogonal axes
of the unit sphere
that is transformed (using chol()) in ellipse3d, so plotting the axes
transformed in the
same way would give me what I want.
Looking at the result returned by ellipse3d, I see a normals component,
but I'm not sure if this
represents what I want, or, if it is, how to use it to draw the ellipse
major axes in the plot.
> e1 <-ellipse3d(cov, centre=mu, level=0.68)
> str(e1)
List of 6
$ vb : num [1:4, 1:386] 4.95 2.64 2.03 1.00 6.74 ...
$ ib : num [1:4, 1:384] 1 195 99 196 51 197 99 195 27 198 ...
$ primitivetype: chr "quad"
$ homogeneous : logi TRUE
$ material : list()
$ normals : num [1:4, 1:386] 0.290 -0.902 -0.320 1.000 0.635 ...
- attr(*, "class")= chr "qmesh3d"
The normals component contains the surface normals. It is used to help
in rendering the surface, but isn't much use for your purposes.
Unfortunately, I'm not familiar enough with the internals of princomp to
tell you how to get the axes you want.
Duncan Murdoch
-Michael
______________________________________________
[email protected] 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.
