[R] R-0.64.2 vs Spls 5.0

[Thomas Lumley]

On Fri, 2 Jul 1999, Agustin Lobo wrote:

> Dear R makers and users:
> 
> After reading the FAQ and the comments on the
> comparison between R and S, I still have an important
> (at least for me) question: How is R compared
> to the new version of Splus 5.0  (for unix including Linux)?
> 
> I must pay particular atention to the efficiency
> with large datasets and, more generally speaking,
> to the efficiency with memory management.

There is not a lot known about the precise circumstances that affect the
relative performance of R and S-PLUS.  Tests based on code from "Modern
Applied Statistics with S-PLUS" by Venables and Ripley show R as somewhat
slower. A number of people (including me) have found R to be substantially
faster on specific simulation problems, sometimes involving large datasets
and sometimes not. This is by no means universally true, and we don't have
good rules for predicting when R is faster.  Basically, R copies things
less often but doesn't recover memory particularly efficiently.

If your object size is a substantial fraction of available memory R will
not perform well. S-PLUS may well be better, but it won't be very good
either. 

There are a number of planned changes to the internals of R that should
provide some speedup (better garbage collection, hashing of all
environments), but we don't know how much.

> I use Splus since version 3.x (on Unix) and now
> use Splus 4.5 for Windows (which gives me tons
> of memory problems). I'm considering whether
> going on with Splus and get Splus 5.0 for Linux
> or switch to R or to X-ploRe. 

If you already have substantial experience with S-PLUS 3.x then it would
be relatively inexpensive in time and effort to try R and see if its
performance is satisfactory on your problems. 

If there are strong cost constraints and memory size is an important
barrier it may be worth using R and buying extra memory, which is
comparatively cheap for Intel machines.

Also, you say that you work with spectral analysis of time series. You
would want to make sure that R supports the analyses you want to do, as
its coverage of time series is not as good as some other areas. 

I have no experience at all with XploRe and can't comment on its
performance. 


Thomas Lumley
Assistant Professor, Biostatistics
University of Washington, Seattle



-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-help-request at stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._