[R-gui] Re: R-SIG-GUI digest, Vol 1 #18 - 5 msgs

[A.J. Rossini]

>>>>> "peter" == Peter Dalgaard <BSA <p.dalgaard@biostat.ku.dk>> writes:

    peter> I do actually think it is essential. R is a modular design, and it
    peter> isn't possible to define in advance what a package might wish to
    peter> contain in terms of user interface. Certain things might be boiled
    peter> down to a static description that could sit in an external file or in
    peter> an R data structure, but not all. How would a dynamic graphics
    peter> package like ggobi (um, tkgobi?) fit in for instance? Programmability
    peter> has been *the* key to the success of the S languages, so why should we
    peter> discard the possibility of making our user interfaces programmable?

This last example is one I'm actually working on (gosh, at least I
wish I was, maybe after I finish up the 2 clinical trials protocols on
my desk tommorow...).

So, we've got a great dynamic graphics interface (modulo stability,
speed, and efficiency of Java/SJava/R interaction, ) using Orca.  It
works, all views are linked nicely, you can brush any of the views and
this propogates between views and R (i.e. 
        result <- GrandTour(myDataFrame)
        result$brush(ThisColor,ArrayOfCases)
        CasesColored <- result$select()
works, for about 5 types of high-dimensional views).

BUT...

While I can piece together these views fairly easily, even by building
from within R, it would be far simpler and potentially more flexible
to use a config file to describe the multi-view layout, and linkages
(or lack thereof) between projections.  

Right now, things are hardcoded, which is fine for prototyping and
research, but lousy for real work.

So example #2, which I won't spell out, is: How should one describe a
multiview visualization?  (components (which exist NOW!)  might be
high-dimensional covariates, correlated data, spatial-temporal data,
multivariate timeseries, projection engines (regression, many
shapes/forms) and dynamic projection engines (grand tours, guided and
unguided, possibly mixed w/ regression), compositional data, graph
data.

So, another use-case to ponder.  We've got a alpha-quality
"visualization builder" (stolen UML graph modeling code), and part of
the reason it is alpha is that the structure is a bit different
(forced rooted and highly branched tree, a bit more complex that the
t-test GUI example).

best,
-tony

-- 
A.J. Rossini				Rsrch. Asst. Prof. of Biostatistics
U. of Washington Biostatistics		rossini@u.washington.edu	
FHCRC/SCHARP/HIV Vaccine Trials Net	rossini@scharp.org
-------------- http://software.biostat.washington.edu/ --------------
FHCRC: M: 206-667-7025 (fax=4812)|Voicemail is pretty sketchy/use Email
UW:   Th: 206-543-1044 (fax=3286)|Change last 4 digits of phone to FAX
(my tuesday/wednesday/friday locations are completely unpredictable.)