On Thu, Mar 1, 2012 at 7:53 AM, Robert Dodier <robert.dodier at gmail.com>wrote:
> On 2/29/12, Mixon, Wilson <wmixon at berry.edu> wrote:
>
> > It appears that the simple_linear_regression routine in stats works for
> n <
> > 345.
> > The following works, but n > 344 results in this message: gamma:
> overflow
> > in GAMMAFLOAT.
>
> Looks like the error originates here:
>
> quantile_chi2 (0.975, 343);
> => error
>
> The problem appears to be in IIGAMMA (share/distrib/numdistrib.lisp)
> which computes the inverse of the incomplete gamma function.
> Probably there is no simple way to resolve the problem; I'm guessing
> IIGAMMA will have to apply some different algorithm for large arguments.
>
>
Looks like the actual problem is in gamma_incomplete_regularized. The
current method just computes gamma_incomplete and then divides by gamma.
This kind of defeats the point of the regularized function. We need to
come up with a different function that computes
gamma_incomplete_regularized without having to call gamma for large
arguments.
Ray.