Version française
Home     About     Download     Resources     Contact us    

The Caml Hump: Hansei

Hansei

HANSEI is the the embedded domain-specific language for probabilistic programming: for writing potentially infinite discrete-distribution models and performing exact inference, importance sampling and inference of inference. HANSEI is an ordinary OCaml library, with probability distributions represented as ordinary OCaml programs. Delimited continuations let us reify non-deterministic programs as lazy search trees, which we may then traverse, explore, or sample. Thus an inference procedure and a model invoke each other as co-routines. Thanks to the delimited control, deterministic expressions look exactly like ordinary OCaml expressions, and are evaluated as such, without any overhead. Inference procedures and probabilistic models are both ordinary OCaml functions. Both may be defined and extended by library users; both may appear in each other's code. Performing inference on a model that includes calls to inference procedures lets us parameterize distributions by distributions, and lets inference procedures measure their own accuracy. One application is modeling agents reasoning about each other's limited reasoning.
Homepage ]
Author:Chung-chieh Shan and Oleg Kiselyov.
Last modification date:02-Jun-2009
Development status:Development code
Kind: Libraries :: Native OCaml libraries
Topic: Science :: Maths and Logic
Homepage:http://okmij.org/ftp/kakuritu/