Direct Graphical Models
v.1.5.2

Here we introduce the application of the DGM library to some wellknown probabilistic models. Particaulary, we show how to compose graphical models with the help of DirectGraphicalModels::CGraph class. We fill the model by hand with potentials and study inferece and decoding processes on them. First we demonstrate bruteforce algorithms for exact decoding and inference on a small graph, and consider a possible use of marginal probabilities calculated by inference algorithm for approximate decoding. Next we show messagepassing algorithms for approximate inference, and consider the difference between sumproduct and maxsum approaches. Finally, we consider sumproduct messagepassing algorithms, which provide an efficient framework for exact inference in chain and treestructuresd graphs.