[R] Attempt to customise the "plotpc()" function

Nikos Alexandris nikos.alexandris at felis.uni-freiburg.de
Sun May 16 20:37:53 CEST 2010


Peter Ehlers wrote:
> Nikos,
> 
> I think you can just replace the line
> 
>       pc <- princomp(x[,1:2], scores=TRUE, na.action=na.fail)
> 
> with
> 
>      pc <- prcomp(x[,1:2], retx=TRUE, center=pc.center,
>                   scale.=pc.scale, na.action=na.fail)
> 
> and rename the components of pc
> 
>      names(pc) <- c('sdev', 'loadings', 'center', 'scale', 'scores')

Right. Υet, it is still not enough. I had to change the definition of the 
limits that feed viewport mainly because of the huge difference of an unscaled 
vs. scaled dataset before the pc-analysis takes place.

Because I want to give (me) the option to have really informative plots, I've 
added an extra grid.points() in case the data are transformed (centered and/or 
scaled) to print both the original and the transformed (with another pch 
and/or color) point cloud.
 
> and then use the rest of the plotpc() code as is (except for
> maybe having to use flip1=TRUE, etc).

Hmm... I am _now_ working on it to understand how I could make this 
"automatic"!.

If I give flip1, flip2 (=TRUE) the histograms are located where they should 
(optically) be printed but the text (rotation angle) that accompanies the 
histogram is I think not correct. It is quite the opposite angle that is being 
printed.

Any ideas?
 
> As to why other functions used in plotpc() are not printed
> when you ask R to print plotpc(): why should they be? Can you
> imagine the mess that would result if you got the printouts of
> is.na(), pushViewport, popViewport, ...? Egad!

Thank you Peter. I understand it now. [ Ignorant me but if you don't know 
something you will probably do mistakes (which is after all the learning 
process. ]
 
> Anyway, as you've discovered, when you want to modify code, look
> at the sources.

Thank you Peter. Kindest regards, Nikos



More information about the R-help mailing list