Dear R-list,
Among the (R-)tools, I've seen on the net, for (bivariate) Principal
Component
scatter plots (+histograms), "plotpc" [1] is the one I like most.
By default it performs PCA on a bivariate dataset based on R's
"princomp()"
(which is the eigenvector-based algebraic solution to PCA). I would like to
modify "plotpc()" in order be able, as an end-user, to define whether
the PCA
should be performed by using:
(a) "princomp()" (the default currently)
(b) the "prcomp()" function (which is the SVD-based algebraic
solution to
PCA)
Of course I want to have control over the parameters that "prcomp()"
accepts,
namely:
- data centering (argument "center")
- data scaling (argument "scale")
Hopefully this is the right place to post this question(s). I started the
modification by attempting first to get a "prcomp" version of
"plotpc()"
(named it "plotpc.svd()") by altering the following:
- accept the arguments "pccenter" and "pcscale" in line 1:
--%<---
"function ( x , pccenter = TRUE, scale = FALSE, xrange =... )"
-->%---
- accept both the "center" and the "scale" parameters for
the "scale()"
function in line 134:
--%<---
x <- scale (x , center = pccenter , scale = pcscale )
-->%---
- change "princomp" to "prcomp" in line 140 as well as all
required
arguments (that I know off):
--%<---
pc <- prcomp ( x[, 1:2] , center = FALSE , scale = FALSE , retx = TRUE ,
na.action = na.fail )
-->%---
-- in line 144:
--%<---
angle <- atan2(pc$rotation[2, 1], pc$rotation[1, 1])
-->%---
-- in lines 147 and 148, I replaced "pc$scores" with
"pc$x"
Running the modified function ends up with an error:
--%<---
# "x" is a data.frame with two columns and works fine with plotpc()
plotpc.svd (x , pccenter = TRUE , pcscale = FALSE )
Error in inherits(unit, "unit") : could not find function
"un"
# getting more info
traceback()
5: inherits(unit, "unit")
4: is.unit(y)
3: textGrob(label = label, x = x, y = y, just = just, hjust = hjust,
vjust = vjust, rot = rot, check.overlap = check.overlap,
default.units = default.units, name = name, gp = gp, vp = vp)
2: grid.text(main, y = un(ymean + xlim[2] + xrange/10), gp = gpar(cex = 1,
font = 2))
1: plotpc.svd(x, pccenter = TRUE, pcscale = FALSE)
-->%---
I am bit lost now about where I should continue looking for required
modifications in the code. Any hints?
Thanks in advance, Nikos
On Saturday 15 of May 2010 19:09:30 Nikos Alexandris wrote:> Among the (R-)tools, I've seen on the net, for (bivariate) Principal > Component scatter plots (+histograms), "plotpc" [1] is the one I like > most.[...] [1] <http://cran.r-project.org/web/packages/plotpc/index.html>
Nikos Alexandris:> Among the (R-)tools, I've seen on the net, for (bivariate) Principal > Component scatter plots (+histograms), "plotpc" [1] is the one I like > most.[...]> I started the modification by attempting first to get a "prcomp" version of > "plotpc()" (named it "plotpc.svd()") by altering the following:[...]> I am bit lost now about where I should continue looking for required > modifications in the code. Any hints?Once again I am replying to myself ;-) I've spend many-many hours of searching in manuals, on the net and trying crazy things to understand where this mystical (to me) function "un()" could be sourced from. Eventually I decided to change all of its occurrences with "unit()" as I am sure this was the function meant to be used but I hit another wall, another unknown function my.plot.something(). I was then sure that they are "sourced" from somewhere. At some point I was enlightened and had a look in the source "plotpc.R" where I just found what I was expecting to found ;-). It may look silly but to the unexperienced useR it is not. I was only looking at the print-out of "plotpc" (without the parentheses) and was puzzled that the "un()" function is nowhere. - Question: why isn't the whole source of plotpc.R printed out with "plotpc"? Or why isn't any clue given where this "un()" function is coming from? "methods()" and "getAnywhere()" say anything about it. * Note to self: look at the Source.R Anyhow, it was a long learning-night-session and I am good to go. In fact, I am very close to what I want to do. I've added the option to use either "prcomp" or "princomp" as well as if the input dataset should be centered and/or scaled. I have still some strange issue with respect to how the plotted histograms (of PC1 and PC2) along with their text are flipped. Hopefully I'll fix this too. I will be contacting soon the author to post him my modifications as enhancement wishes. If anybody is interested to know or help please post here. Regards, Nikos