Bayesian Decision-Making toolbox for C++

1

Version:

0.1

Author:

Vaclav Smidl

BDM is a collection of methods for selected tasks of Bayesian decision-making, such as estimation, filtering and control.

Features

At present the following algorithms are implemented:

• Bayesian filtering : Kalman filter, EKF, patricle filter (PF),
  • these can be combined mutualy together in mode demanding schemes, see marginalized particle filter MPF

• Classification using mixtures of exponential famiuly models (MixEF),
• Density estimation : using mixtures (MixEF), density composition (merger)

Download and Use

The library is available under GPL, see installation instructions on page How to install BDM

Precompiled Mex files for use within Matlab are available How to write and use mex files for Matlab

Design Approach

The toolbox is designed using object-oriented approach with two design criteria:

• relation to mathematics,
• efficient evaluation, while the first one is more important than the latter.

Hence, each mathematical object such as probability density is represented by one software object. The resulting algorithms are then implemented as operations on these objects. In cases when more efficient solution can be achived when this structure is not respected, a parallel implementation is created and clearly marked as specific.

OpenMP is used to achive efficient implementation on parallel architectures.

Implementation

BDM is build on top of IT++ which wraps numerically efficient operations of linear algebra into easy to use C++ classes. Thanks to this excellent library, writing of numerical algorithms is as easy as in Matlab but we gain significant advantages:

• computational speed comparable to built-in Matlab function, and surpassing interpreted Matlab in order of magnitudes,
• native support for object-oriented programming,
• support for templates which is often more appropriate than object-oriented programming,
• cross-platform compatibility.