
Boost : 
From: Neal Becker (ndbecker2_at_[hidden])
Date: 20071226 08:13:56
John Maddock wrote:
> Joel Eidsath wrote:
>> Paul A. Bristow wrote:
>>> This looks too obviously lifted from NR in C (or FORTRAN) to fit the
>>> Boost license.
>>
>> Yes, as I mentioned, it's from a NR FORTRAN recipe. I don't know if
>> the conversion from FORTRAN to C++ makes it a nonderivative work. The
>> code certainly wouldn't work in poisson_distribution.hpp as is.
>
> A direct or mechanical conversion almost certainly does make it a derived
> work of the NR original: the best way is to go back to the original
> reference if possible and work from there. One recent reference is:
>
> Ahrens, J.H. and Dieter, U. (1982).
> Computer generation of Poisson deviates from modified normal
> distributions. ACM Trans. Math. Software 8, 163179.
>
>>> But the algorithm should still be usable? (But I am not qualified to
>>> judge).
>>>
>>
>> If not, then at least the fix (in the original part of my post)
>> should be used.
>>
>> Right now, using [random] poisson_distribution is slow and dangerous
>> for any mean over 750. If it is changed to use ln instead of exp, it
>> will only be slow. That is completely unrelated to the NR code later
>> on in my post, which is a (serious) speed optimization.
>>
>>> ln gamma (and poisson) is available from the Math Toolkit, but this
>>> is a random requirement and you may value speed more than accuracy?
>>
>> I don't see any way to use the Math Toolkit poisson distribution code
>> to actually generate poisson random variables.
>
> You *could* apply the poisson quantile to a uniformly distributed variable
> to obtain a
> poisson one, but I wouldn't honestly recommend it for that purpose :/
>
>
> I'll look into the Poisson PDF issue,
>
You may find this useful:
http://statistik.wuwien.ac.at/unuran
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk