DBNL is a cross-platform library that offers a variety of implementations of Bayesian networks and machine learning algorithms.
It is a flexible library that covers all aspects of Bayesian netwoks from representation to reasoning and learning. It allows you to create simple static networks as well as complex temporal models with changing structure.
It can handle highly non-linear dependencies between multivariate random variables. The particle based inference can answer arbitrary questions given the provided evidence and can even cope with multimodal densities. The library supports the most common types of densities and conditional densities, like uniform or normal densities and facilitates user defined density functions.
To enable easy use the library is taking account of modern development techniques like policy based design and template programming.
All these properties make it applicaple for a wide range of applications.
Features
- Bayesian Networks
- Dynamic Bayesian Networks
- Approximate particle inference
- User defined conditional densities
- Multi modal distributions
- Mixed continous, discrete state spaces
- Non-linear models
- Machine learning algorithms
- Visualization
- Serialization