The Libra machine learning toolkit includes implementations of a variety of algorithms for learning and inference with Bayesian networks and arithmetic circuits:

Learning algorithms -- Structure learning for BNs and ACs; Chow-Liu algorithm; AC weight learning

Inference algorithms - Mean field, belief propagation, Gibbs sampling, AC variable elimination, AC exact inference

Libra's strength is exploiting context-specific independence (such as decision tree CPDs) to allow exact inference in models with high treewidth.