In this section, we will introduce two tools in Java that are very popular for probabilistic graph modeling.
OpenMarkov is a Java-based tool for PGMs and here is the description from www.openmarkov.org:
OpenMarkov is a software tool for probabilistic graphical models (PGMs) developed by the Research Centre for Intelligent Decision-Support Systems of the UNED in Madrid, Spain.
It has been designed for: editing and evaluating several types of PGMs, such as Bayesian networks, influence diagrams, factored Markov models, and so on, learning Bayesian networks from data interactively, and cost-effectiveness analysis.
OpenMarkov is very good in performing interactive and automated learning from the data. It has capabilities to preprocess the data (discretization using frequency and value) and perform structure and parameter learning using a few search algorithms such as search-based Hill Climbing and score-based PC. OpenMarkov stores the models in a format known as pgmx. To apply the models in most traditional packages there may be a need to convert the pgmx models to XMLBIF format. Various open source tools provide these conversions.
Here we have some screenshots illustrating the usage of OpenMarkov to learn the structure and parameters from the data.
In Figure 20, we see the screen for interactive learning where you select the data file and algorithm to use:
The next step is the Preprocessing tab (Figure 21) where we can select how discretization is done:
Finally, in Figure 22, we see the display of the learned Bayes network structure:
Weka's Bayes Network editor for interactive and automated learning has a large number of options for Bayes network representation, inference and learning as compared to OpenMarkov. The advantage in using Weka is the availability of a number of well-integrated preprocessing and transformation filters, algorithms, evaluation, and experimental metrics.
In Figure 23, we see the Bayes Network Editor where the search algorithm is selected and various options can be configured:
The learned structure and parameters of the BayesNet are shown in the output screen in Figure 24:
3.149.250.11