wolffd@0: 
wolffd@0: <h2><a name="features">Major features</h2> 
wolffd@0: <ul> 
wolffd@0:  
wolffd@0: <li> BNT supports many types of
wolffd@0: <b>conditional probability distributions</b> (nodes),
wolffd@0: and it is easy to add more.
wolffd@0: <ul> 
wolffd@0: <li>Tabular (multinomial)
wolffd@0: <li>Gaussian
wolffd@0: <li>Softmax (logistic/ sigmoid)
wolffd@0: <li>Multi-layer perceptron (neural network)
wolffd@0: <li>Noisy-or
wolffd@0: <li>Deterministic
wolffd@0: </ul> 
wolffd@0: <p> 
wolffd@0:  
wolffd@0: <li> BNT supports <b>decision and utility nodes</b>, as well as chance
wolffd@0: nodes,
wolffd@0: i.e., influence diagrams as well as Bayes nets.
wolffd@0: <p> 
wolffd@0:  
wolffd@0: <li> BNT supports static and dynamic BNs (useful for modelling dynamical systems
wolffd@0: and sequence data).
wolffd@0: <p> 
wolffd@0:  
wolffd@0: <li> BNT supports many different <b>inference algorithms</b>,
wolffd@0: and it is easy to add more.
wolffd@0:  
wolffd@0: <ul> 
wolffd@0: <li> Exact inference for static BNs:
wolffd@0: <ul> 
wolffd@0: <li>junction tree
wolffd@0: <li>variable elimination
wolffd@0: <li>brute force enumeration (for discrete nets)
wolffd@0: <li>linear algebra (for Gaussian nets)
wolffd@0: <li>Pearl's algorithm (for polytrees)
wolffd@0: <li>quickscore (for QMR)
wolffd@0: </ul> 
wolffd@0:  
wolffd@0: <p> 
wolffd@0: <li> Approximate inference for static BNs:
wolffd@0: <ul> 
wolffd@0: <li>likelihood weighting
wolffd@0: <li> Gibbs sampling
wolffd@0: <li>loopy belief propagation
wolffd@0: </ul> 
wolffd@0:  
wolffd@0: <p> 
wolffd@0: <li> Exact inference for DBNs:
wolffd@0: <ul> 
wolffd@0: <li>junction tree
wolffd@0: <li>frontier algorithm
wolffd@0: <li>forwards-backwards (for HMMs)
wolffd@0: <li>Kalman-RTS (for LDSs)
wolffd@0: </ul> 
wolffd@0:  
wolffd@0: <p> 
wolffd@0: <li> Approximate inference for DBNs:
wolffd@0: <ul> 
wolffd@0: <li>Boyen-Koller
wolffd@0: <li>factored-frontier/loopy belief propagation
wolffd@0: </ul> 
wolffd@0:  
wolffd@0: </ul> 
wolffd@0: <p> 
wolffd@0:  
wolffd@0: <li> 
wolffd@0: BNT supports several methods for <b>parameter learning</b>,
wolffd@0: and it is easy to add more.
wolffd@0: <ul> 
wolffd@0:  
wolffd@0: <li> Batch MLE/MAP parameter learning using EM.
wolffd@0: (Each node type has its own M method, e.g. softmax nodes use IRLS,<br> 
wolffd@0: and each inference engine has its own E method, so the code is fully modular.)
wolffd@0:  
wolffd@0: <li> Sequential/batch Bayesian parameter learning (for fully observed tabular nodes only).
wolffd@0: </ul> 
wolffd@0:  
wolffd@0:  
wolffd@0: <p> 
wolffd@0: <li> 
wolffd@0: BNT supports several methods for <b>regularization</b>,
wolffd@0: and it is easy to add more.
wolffd@0: <ul> 
wolffd@0: <li> Any node can have its parameters clamped (made non-adjustable).
wolffd@0: <li> Any set of compatible nodes can have their parameters tied (c.f.,
wolffd@0: weight sharing in a neural net).
wolffd@0: <li> Some node types (e.g., tabular) supports priors for MAP estimation.
wolffd@0: <li> Gaussian covariance matrices can be declared full or diagonal, and can
wolffd@0: be tied across states of their discrete parents (if any).
wolffd@0: </ul> 
wolffd@0:  
wolffd@0: <p> 
wolffd@0: <li> 
wolffd@0: BNT supports several methods for <b>structure learning</b>,
wolffd@0: and it is easy to add more.
wolffd@0: <ul> 
wolffd@0:  
wolffd@0: <li> Bayesian structure learning,
wolffd@0: using MCMC or local search (for fully observed tabular nodes only).
wolffd@0:  
wolffd@0: <li> Constraint-based structure learning (IC/PC and IC*/FCI).
wolffd@0: </ul> 
wolffd@0:  
wolffd@0:  
wolffd@0: <p> 
wolffd@0: <li> The source code is extensively documented, object-oriented, and free, making it
wolffd@0: an excellent tool for teaching, research and rapid prototyping.
wolffd@0:  
wolffd@0: </ul> 
wolffd@0:  
wolffd@0: