Crate bayes

This create offers composable abstractions to build probabilistic models and inference algorithms operating on those models.

The trait Distribution offer the basic random sampling and calculation of summary statistic functionality for the typical parametric distributions. Implementors of this trait are located at the distr module.

The trait Estimator offer the fit method, which is implemented by the distributions themselves (conjugate inference) and by generic estimation algorithms. Two algorithms will be provided: ExpectMax (expectation maximization) which returns a gaussian approximation for each node of a generic distribution graph; and Metropolis (Metropolis-Hastings posterior sampler) which returns a non-parametric marginal histogram for each node.

Modules

basis	Basis transformations. model non-linear processes; Moslty via bindings to GSL and MKL (work in progress).
decision	Supports the derivation of optimized decision rules based on comparison of posterior log-probabilities (work in progress).
distr	Traits and implementations for exponential-family probability distributions with support for sampling, summary statistics, and conditioning.
gsl	Auto-generated bindings to GSL (mostly for optimization and sampling).
optim	Algorithm for approximating posteriors with multivariate normals (Expectation Maximization; work in progress).
sim	Full posterior estimation via simulation (Metropolis-Hastings algorithm) and related non-parametric distribution representation (work in progress).