jHMM is a library and application for handling Hidden Markov Models that is written in Java. The models are encoded in a simple XML format that allows you to have arbitrary states, transitions, and alphabets. The library implements sequence state predictions and Baum-Welch training of the model parameters.

Building

Build the code with ant: cd hmm ant jar

Running

By default, the jar file is setup to generate a prediction from a model and input sequence:

java -jar dist/jhmm.jar model.xml input.txt

You can run a test by doing the following:

cd hmm-data java -jar ../hmm/dist/jhmm.jar ../models/two-dice-model.xml dice.txt

and you get the output:

1111111111111111111111111111111111111112222222222222222222222222222111111111111111111111111111111111 Score = -178.57415471540796

You can also run several other programs using the jhmm.sh script:

cd hmm ./jhmm.sh com.milowski.hmm.tools.Forward ../models/two-dice-model.xml ../hmm-data/dice.txt

The following tools are available:

• com.milowski.hmm.tools.Forward - the forward algorithm output
• com.milowski.hmm.tools.Generate - generates random output according to your model
• com.milowski.hmm.tools.ModelCheck - check the syntax of your model
• com.milowski.hmm.tools.Predict - calculate the most likely state for a sequence
• com.milowski.hmm.tools.Train - train a model on a set of sequences