binomial_tpe

[James Theiler <jt at lanl dot gov>]

Thanks, that is an interesting approach.

The code I was talking about is available in GSL as 
gsl_ran_binomial_tpe(rng,p,n), and the source code
is in the file randist/binomial_tpe.c

by the way, i noticed that this is not mentioned
in the revised second edition of the manual; i think
users might like to know that we have a much faster
version available.  (actually, what i think is that
the btpe version should be the standard.)

jt

On Wed, 17 Aug 2005, Chris Paulse wrote:

] This may interest you:
] http://statistik.wu-wien.ac.at/unuran/index.html
] 
] chris
] 
] -----Original Message-----
] From: James P. Theiler [mailto:jtheiler@lanl.gov] 
] Sent: Thursday, August 11, 2005 1:57 AM
] To: Chris Paulse
] Subject: Re: GSL suggestion
] 
] Hi Chris,
]   and thanks for your comments.  In fact, I noticed that the standard GSL
] implementation for binomial variates was way slow, and incorporated what
] I think is just the scheme you describe.  In fact, I corresponded with
] both of the authors of that paper, and started to work on a faster
] Poisson distribution as well.  But Brian decided to put the new version
] in with a different API, I can't remember what it is exactly.  (I'm
] basing all this on memory, since I'm on travel and don't have access to
] the details.)  I should  double-check on that and get back to you with
] more specifics.
] 
] thanks again,
] jt
] 
] 
] > Hi James,
] >
] > I noticed that the c library for low level probability functions uses a
] > different (newer) method for drawing from a binomial distribution than the
] > GSL.  The relevant reference for the newer implementation is
] >
] >
] >
] > Kachitvichyanukul, V. and Schmeiser, B. W. (1988).
] >
] > Binomial random variate generation.
] >
] > Communications of the ACM 31, p216.
] >
] > (Algorithm BTPEC).
] >
] >
] >
] > The implementation of this is in the file rbinom.c in the source
] > distribution for the R package (www.r-project.org
] > <http://www.r-project.org/> ).
] >
] >
] >
] > Chris Paulse
] >
] >
] 
] 

-- 
James Theiler            Space and Remote Sensing Sciences
MS-B244, ISR-2, LANL     Los Alamos National Laboratory
Los Alamos, NM 87545     http://nis-www.lanl.gov/~jt