[Caml-list] ANN: mcmc-ocaml: A Library for Markov-Chain Monte Carlo Computations in OCaml

[Will M Farr <w-farr@northwestern.edu>]

Hello all,

I'd like to announce mcmc-ocaml, a library for Markov-chain Monte Carlo 
computations in OCaml.  You can obtain the library from github at

https://github.com/farr/mcmc-ocaml

It is released under the GPL, version 3.0.  For more information on 
MCMC, see http://en.wikipedia.org/wiki/Markov_chain_Monte_Carlo .

Some highlights of the mcmc-ocaml library:

* mcmc-ocaml provides code to perform MCMC analysis in arbitrary 
parameter spaces.  Parameter values need not be numerical, but can 
instead be any OCaml value: algebraic data types, classes, etc.

* mcmc-ocaml provides useful higher-order functions to combine multiple 
jump proposals with arbitrary weights into a single proposal for 
Metropolis-Hastings sampling, making it easy to have many special jump 
proposals tuned to your specific parameter space structure.

* mcmc-ocaml provides code to read and write MCMC samples from/to files; 
this can be combined with the library of statistical functions or 
user-written code to perform post-processing of MCMC output either in a 
stand-alone program or from the REPL.

* mcmc-ocaml provides many utility functions for performing model 
selection via Reversible-Jump MCMC, either between two models or among a 
larger set of models.  This includes an implementation of the method 
described in http://arxiv.org/abs/1104.0984 for facilitating inter-model 
jumps by interpolating posteriors using single model parameter samples.

* mcmc-ocaml provides several different algorithms for computing the 
Bayesian evidence for a model using MCMC samples from its posterior 
distribution, including the "direct integration" algorithms described in 
http://arxiv.org/abs/0911.1777 .

* mcmc-ocaml comes with complete ocamldoc documentation for its 
functions and types.

mcmc-ocaml is extensively tested, and has been used to constrain the 
mass distribution of stellar-mass black holes in our galaxy 
(http://arxiv.org/abs/1011.1459 ---including the 10-way RJMCMC 
comparisons among the models in that paper), among other research 
projects.  Please direct bug reports, feature requests, and comments to 
this address (w-farr@northwestern.edu).

Thanks,
Will