[R] 3d cloud plot with point size reflecting variable value

[Deepayan Sarkar]

On Tue, Mar 31, 2009 at 11:49 AM, Deepayan Sarkar
<deepayan.sarkar at gmail.com> wrote:
> On Tue, Mar 31, 2009 at 8:57 AM, Dieter Menne
> <dieter.menne at menne-biomed.de> wrote:
>> R User R User <ruser2008 <at> googlemail.com> writes:
>>
>>> I'm using the cloud function to plot the iris data as per the document:
>>> http://www.stat.ucl.ac.be/ISpersonnel/lecoutre/stats/fichiers/_gallery.pdf
>>>
>>> I'd like to change the point size to reflect a fourth variable, as done here
>>> http://www.cis.hut.fi/projects/somtoolbox/download/pics2/shotvs2_origdata.png
>>>
>>> Does anybody know how to do this? There doesn't seem to be an option in
>>> cloud()
>>>
>>
>> My first idea was to add
>>  cex = iris$PetalLength,
>>
>> which failed, taking the first value of cex only. In panel.3dscatter
>> I found the line
>>
>> cex <- rep(cex, length.out = n)
>>
>> Which looks like cex is not vectored.
>
> How so? panel.3dscatter() in fact goes to some lengths to be
> vectorized: 'n' is the number of points being plotted, and 'cex' is
> replicated to be equally long here. This is needed so that the 'cex'
> values can later be reordered in the same way as the points being
> plotted (by decreasing depth). The following should work:
>
> cloud(Sepal.Length ~ Petal.Length * Petal.Width, data = iris, pch = 1,
> cex = with(iris, 3 * Sepal.Width / max(Sepal.Width)))
>
> One caveat is that this will not work in a multi-panel plot. In that
> case, a custom panel function will need to create a suitable subset of
> 'cex' (cex[subscripts]) before calling panel.3dscatter().

I think I see the problem: this will also not work if there is a
'groups' argument. In general, that will also require a panel
function, but another option is to use vectorized color:

with(iris,
     cloud(Petal.Length ~ Sepal.Length * Sepal.Width, pch = 16,
           cex = 2 * Petal.Width / max(Petal.Width),
           col = c(2, 3, 4)[as.numeric(Species)], aspect = c(0.7, 2)))

-Deepayan