random numbers according to user defined distribution ??

[Robert Kern]

sturlamolden wrote:
> Alex wrote:
> 
>> I wonder if it is possible in python to produce random numbers
>> according to a user defined distribution?
>> Unfortunately the random module does not contain the distribution I
>> need :-(
> 
> There exist some very general algorithms to generate random numbers
> from arbitrary distributions.
> 
> The most notable of these are "Markov Chain Monte Carlo", e.g. the
> Metropolis-Hastings algorithm. It is very easy to implement in any
> programming language. The nature MCMC algorithms makes it inefficient
> when implemented in pure Python. But you can get tremendous speedup by
> simulating multiple Markov chains in parallel, by means of vectorizing
> with NumPy.
> 
> A relative of Metropolis-Hastings which may also be applicable to your
> problem is pure "rejection sampling". It is far less efficient, but
> produces no autocorrelation in the samples.

I don't know. I think the certainty that rejection sampling actually gives you 
the desired distribution as opposed to MH's uncertainty is very much a 
worthwhile tradeoff for univariate distributions. Sure, you throw away fewer 
samples, but you know that MH isn't throwing away samples that it ought to. 
Personally, I view MH as a last resort when the dimensionality gets too large to 
do anything else. But then, that's just my opinion.

-- 
Robert Kern

"I have come to believe that the whole world is an enigma, a harmless enigma
  that is made terrible by our own mad attempt to interpret it as though it had
  an underlying truth."
   -- Umberto Eco