Robert Dodier <robert.dodier at gmail.com> writes:
> I read Ihaka's paper a while ago (so I apologize for my faulty memory)
> and iirc he is interested in reimplementing R in Lisp in order to
> make it run faster. I was kind of puzzled by the that. Obviously (well, it's
> obvious to me anyway) the big payoff is to combine numerical and symbolic
> computations.
I agree with you that the capability to combine numerical and symbolic
computation is a big pay-off but I also see a clear benefit in having
CLisp as an underlying engine simply to make the loop run faster.
>
> I would be very interested in porting some R capabilities into Maxima.
> I made a half-hearted effort to implement a "data frame" construct.
> I guess I should finish it.
>
> Another statistical package that could benefit from combined
> symbolic & numerical computations is the Bayesian inference
> package BUGS. I made some progress (far from a complete implementation)
> of a scheme to attempt symbolic computation first and then fall
> back on a numerical scheme. (The calculations of interest are integrals.)
> I wrote a paper about that if you're interested.
>
> I'd be interested to hear yours or anyones thoughts about this stuff.
I fully agree on this point (the potential benefit of a software like
Maxima for Bayesian inference). I've myself committed a couple of papers
using the MCMC methodology
(http://intl-jn.physiology.org/cgi/content/abstract/91/6/2910,
http://fr.arxiv.org/abs/q-bio.QM/0405012) and I think that this approach
is really challenging as far as software development is concerned since
you want to leave as much freedom as possible to the user as far as
model specification (likelihood function and priors) is concerned while keeping
run-times reasonable on realistic data sets. As far as efficiency
(run-time) is concerned, clearly doing as much as possible symbolically
should help and being able to generate compiled code without too much
trouble should help when the numerical methods can't be avoided.
Christophe
--
Most people are not natural-born statisticians. Left to our own
devices we are not very good at picking out patterns from a sea of
noisy data. To put it another way, we are all too good at picking out
non-existent patterns that happen to suit our purposes.
Bradley Efron & Robert Tibshirani (1993) An Introduction to the Bootstrap
--
Christophe Pouzat
Laboratoire de physiologie c?r?brale
CNRS UMR 8118
UFR biom?dicale de l'Universit? Paris-Descartes
45, rue des Saints-P?res
75006 PARIS
France
tel: +33 (0)1 42 86 38 28
fax: +33 (0)1 42 86 38 30
mobile: +33 (0)6 62 94 10 34
web: http://www.biomedicale.univ-paris5.fr/physcerv/C_Pouzat.html