Alexander Klein <alexander_klein_uni at web.de> wrote:
	I'm currently looking for a sensible environment to do scientific computing
	for my PhD thesis that I hope to start in October.
	...
	The general idea is to use Squeak for visualization purposes and
	statistics, while the really hard work is done by an external kernel
	written in C or C++.
	
As a Squeak enthusiast, may I urge you to look very seriously at R
before considering Squeak?

- R is an open source implementation of the S programming language for
  statistics.  (See http://www.r-project.org/)
- There are oodles of textbooks about using S.  A good one is
  "Modern Applied Statistics with S" by Venables & Ripley, but there
  are others.
- R has a ton of stuff built in.  Via the "Comprehesive R Archive
  Network" sites (CRAN) (http://cran.r-project.org) you have access to
  more statistical and visualisation packages than you would have dreamed
  possible.
- R runs on pretty much any UNIX, on Windows, and on MacOS X (it used to
  run on MacOS 8 and 9; new versions of R will only be supported on MacOS X).
- There are two kinds of foreign function interface in R; the normal kind is
  where you pass arrays of numbers &c between R and C/C++/Fortran (-77 only,
  as yet); the other kind is where C/C++ code can grovel over R internal
  data structures.
- R is comprehensively documented.  There are tutorials, language reference,
  guide to exchanging data with other systems, guide to using the foreign
  interface, and all the built in functions and classes are *thoroughly*
  documented using a TexInfo-like system (current version comes to >2000
  printed pages).
- When the graphical methods in R are not quite up to your demands (like
  spinning plots, for example), there's an R interface to XGobi (now GGobi).
- The R-help mailing list is extremely active; it is as helpful as the
  Squeak mailing list (no, *more* helpful; you're not going to pick up an
  education in modern statistics from the Squeak mailing list (:-)) and
  (if you discount the BFAV messages) about equally voluminous.
  
Let me put it this way.  I have both Squeak and R.  If I wanted to write
an upper layer for visualisation and statistics over an external C/C++
kernel, I would *certainly* use R rather than Squeak.  R is *designed* to
be "a sensible environment to do scientific computing".

No slur on Squeak is intended, it's just that it's rather easier to
call "glm" in R than to write my own Generalised Linear Model code in Squeak...