Subject: ugly weird expression to be passed to R??
From: Robert Dodier
Date: Thu, 31 Jul 2008 13:29:59 -0600
On Thu, Jul 31, 2008 at 12:28 PM, dlakelan <dlakelan at street-artists.org> wrote:
> I did try using lbfgs to get the maximum likelihood estimates, and it
> did work, even with 800+ data points in the likelihood expression.
I'm glad to hear it worked, but I'm a little confused. Do you mean
the log likelihood is a literal sum of 800 terms? I think it should be
possible to avoid that, and if it's not, I'll fix it so it is.
> I may try doing a MCMC sampling method for getting a bayesian
> posterior sample from the coefficients. Has anyone done anything
> like that in maxima before?
I did something like it (but not MCMC). I think the right way to go about
it is to represent the problem as a Maxima expression, look for any
integrals that can be solved symbolically, and approximate the others
by a suitable method (e.g. MCMC). This is reminiscent of BUGS but
BUGS is strictly numerical. I'll send you a paper I wrote about the
stuff I worked on.
Robert Dodier