I can work on some tests in a few days, though I'm busy at the moment. There should be two tests: one set that tests would be compare the results of the factorial form and the Stirling form where they are in good numerical agreement, basically for 50 < x < 100, and ensure that they agree within bit noise (by one measure or another).
The second set of tests are to ensure that the code for the combined form (called pmf_poisson) gives results that are *identical* to those of the factorial form for x < 50, and the Stirling form for x > 100. David On Wed, Oct 31, 2012 at 2:28 PM, Maggie X <[email protected]> wrote: > Thanks for the update, David! Would you mind adding a couple test cases? > I'd be happy to add it too if you can suggest some numbers to test. > > Best, > Maggie > > On Wed, Oct 31, 2012 at 12:38 PM, David Mertens > <[email protected]>wrote: > >> Maggie (and other Stats PDLers) - >> >> I ran into some numerical stability issues with the Poisson probability >> mass function. The Poisson distribution is mathematically well-behaved, but >> is not numerically stable. Fortunately, I was able to use Stirling's >> approximation to increase the range of accuracy and the allowed input range >> to effectively work for anything. >> >> I've issued a pull request. You can check it out and comment here: >> https://github.com/maggiexyz/PDL-Stats/pull/1 >> >> David >> >> P. S. Maggie's recent switch to Github didn't spur my interest in >> PDL::Stats (blame my research for that), but it sure made it a heck of a >> lot easier for me to manage this proposed modification. :-) >> >> -- >> "Debugging is twice as hard as writing the code in the first place. >> Therefore, if you write the code as cleverly as possible, you are, >> by definition, not smart enough to debug it." -- Brian Kernighan >> >> > -- "Debugging is twice as hard as writing the code in the first place. Therefore, if you write the code as cleverly as possible, you are, by definition, not smart enough to debug it." -- Brian Kernighan
_______________________________________________ Perldl mailing list [email protected] http://mailman.jach.hawaii.edu/mailman/listinfo/perldl
