[tech-spec] Re: R arguments

  • From: Dirk Eddelbuettel <edd@xxxxxxxxxx>
  • To: tech-spec@xxxxxxxxxxxxx
  • Date: Thu, 14 Oct 2004 11:23:18 -0500

On Thu, Oct 14, 2004 at 12:06:32PM -0400, Tom McCubbin wrote:
> Thank you.  I thought my text was invisible.
> I too have a tool (Tom's Big Fat Trader).  The name is not perfectly 
> suitable, as it is really for back testing, simulating the 
> impact/interelationships of models, allocation models, and dynamic 
> universe constraints.   As well as analysis of the results.
> It is written in C++, and uses a system to load modules at runtime that 
> can implement various components of the system(s).  However, it is not 
> accessible to many due to the language, so i'm embedding python into the 
> system to make it scriptable and improve productivity.  I looked at Rpy 
> and was not too keen on its implementation.  So i am considering 
> re-wraping R to make it more friendly from Python.
> Do you have any input on what you would like to see in an enhanced 
> R-wrapper to Python?   I'm happy to contrib the code if it is 
> helpful...i'm a long time open source guy.

Have you seen Duncan Temple Lang's work (formerly Bell Labs / ATT Research,
now at UC Davis) at wwww.omegahat.org?  He has a _ton_ of projects many of
which involve embedding R into other things (databases, browsers, general
toolkits like Gtk, spreadsheets like gnumeric, etc pp ...) as well as links
to other language such as Java, Perl, Python.  He has a genius-level
productivity, but never quite manages to get his software to broad release /
production level. I am Debian maintainer for some of his code, and had many
goes at packaging RSPerl and RSPython but never quite found them
sufficiently developed to do it.

To fully tie Python to R you need a better than average understanding of the
innards of the two systems. That limits the number of contributors and
excludes me. But you could talk to people on the r-devel list (which I read)
or on the omegahat lists (which I don't read, not sure how active theys
still are).

Personally, I changed my viewpoint over the years and now think that R gets
so many things so darn right when it comes to data work, is cross-platform,
robust and easily extensible (with e.g. C++) that I'd make it the core
rather than the satellite. But recognising the uphill struggle that leanring
the S language is, I can see how loosely coupling it may be a better first
step for getting a toe in the water.


Those are my principles, and if you don't like them... well, I have others.
                                                -- Groucho Marx

Other related posts: