Re: cross tables in reports?

[Stanley A. Klein]

I took a look at the links on the R homepage and found something called
RSPython that appears to provide interfaces both ways, i.e., R called from
Python and Python called from R.  It looks like you need to know something
about the objects involved, so it's an interface that needs to be
programmed for specific purposes.

The easiest way for GNUe to quickly do an interface (if desired) would be
to write files using Reports in a format that can be read by R, use the
Python to R interface to send R commands to read the input files and
perform various functions, and then write the output from R to a file that
could either be printed or imported to a database.  It could also be done
directly using the databases via the database interfaces that Gontran
discusses.

I would tend to try to do the Python-to-R calls through something that is
user-definable and not a basic part of the GNUe system.  That means R does
not become a GNUe basic dependency unless someone wants to do heavy-duty
statistical stuff.  If I understand the concept of GNUe triggers and
business logic (and I might not :-), I think trigger or business logic code
would be the place to put the R interface, importing as necessary the
approriate components from RSPython.

Of course, one simple option is to write the Reports output and then call
up R separately to do the analysis.

BTW, there appears to be work ongoing on GUI interfaces for R, but it
doesn't seem to be very far along.  It might be nice to let the R folks
know that GNUe might be a good candidate for that.

Also, in terms of "wacky statistics thingees" there are lots of them that
might be of interest to a business strategic planner or market researcher.
It's good to know that there is a top quality statistics package available
as free software.


Stan Klein


At 06:36 AM 11/2/2002 -0500, Gontran Zepeda <address@hidden> wrote:


>It so happens I have spent part of the past couple weeks hacking up a kludgey
>perl *cough* package that interacts with R by providing simple routines to
>reliably manipulate input and output to R which is executed via system calls.
>So after seeing Stanley's post about how to specifically go about pivoting
>data in R I thought I'd pipe up.
>
>First, R is difficult to work with in that it's syntax is, I guess, very
>similar to S and quite un-gnu like (hard to figure out how to do simple
>things from within R).  However, R is excellent for statistics.
>
>Second, R doesn't have a python API that I'm aware of.  There is however an
>API for using python from _within R_ at http://www.omegahat.org, which is
>something I guess, but rather seems the wrong direction.  Most extension of R
>seems to be unsurprisingly R-centric.
>
>What R _does_ have (thanks again apparently to the guys at omegahat) is a DBI
>type interface.  That's right kids, get data from mysql, postgresql or plain
>flat files into your analysis space directly from R.  I've used this with
>success from within my R wrapper package.
>
>An improved version of this wrapper in implemented in python could be done,
>but I believe that providing number analysis facilities through R should be
>optional at best (yet another dependency).  On the other hand, there's the
>power of R.
>
>Doesn't jamest have pivot tables working anyway?  Are there other wacky
>statistics thingees that need to be done that R could do conveniently?
>