[SciPy-dev] Warning about remaining issues in stats.distributions ?

Yaroslav Halchenko lists@onerussian....
Tue Dec 9 16:18:45 CST 2008


> In my bugfixes, I temporarily removed the rdist._cdf, since the
> generic method works also for large parameters, while
> special.hyp2f1(0.5,1.0-c/2.0,1.5,x*x) does not work for large c. Once

I can only confirm that, and that is why I had to manually redefine
rdist_gen in the same fashion to be used with elderly scipy 0.5.2-0.1
(with def cdf being removed, and also rdist.veccdf.nin = 2),
otherwise I was getting arbitrary numbers in the output of cdf for some
large values of c (10000) and various values of x in the tails.

> I know over which domain the special functions are reliable,
> dispatching to the generic methods only for some part of the parameter
> space will be an improvement, but this requires some time consuming
> testing.

aren't CPU time is relatively cheap these days? Creating a simple test
battery which would sample the space of arguments with emphasis
on boundaries might prove to be informative to hunt down buried
problems. Alternatively, indeed, it is possible just to wait for someone
to hit the issue on his data / favorite distribution ;-)


-- 
                                  .-.
=------------------------------   /v\  ----------------------------=
Keep in touch                    // \\     (yoh@|www.)onerussian.com
Yaroslav Halchenko              /(   )\               ICQ#: 60653192
                   Linux User    ^^-^^    [175555]




More information about the Scipy-dev mailing list