HmmLib is a Java class library which contains all essential functionalities of hidden Markov models. It is not intended to be used for commercial purposes, for the efficiency was not the primary factor in its implementation. Rather, it is hoped that the implementation is clear and transparent enough that it can be used as a model program for students wishing to learn basics of HMMs. Please visit its home page for more information.
You need Java Platform 2 Development Kit (JDK2) in order to be able to use this library. JDK2 is available only on Windows and Solaris as of this writing. Please visit Sun's Java Site for further information. You also need XML parsers in Java. Please see below.
jar xvf hmmlib03.jar
docs/ | Documentation directory |
src/ | Source directory |
hmmlib.jar | Class files. Do not unzip this archive. |
hmm03.dtd | Document type definition file for HmmLib file format. |
readme.html | This file. |
MPL-1_0.html | Mozilla public license. |
set CLASSPATH=%CLASSPATH%;"YOUR_DIRECTORY"\hmmlib.jar
HmmLib contains two essential classes and some utility classes. Model.java is a core class which stores all model parameters relevant to the underlying Markov model, whereas Hmm.java implements all essential functions of a hidden Markov model, including Forward-backward, Viterbi, and Baum-Welch algorithms.
XML parsers of Sun and IBM are also needed from HmmLib v 0.3, since it uses XML as its internal file format to save HMM model parameters. If you decide not to use XML, or if you have trouble running XML parsers, you can save your model parameters as ASCII formats. The source files are heavily commented and you can easily find different options for load and save functions.
Also, you can use different XML parsers in your program. Please modify necessary com.bluecraft.hmm.util.* classes as needed. In that case, don't forget to change the package name, since com.bluecraft.* is the namespace I'm using. If you don't have XML parsers, you can download them from the Download Page.
Please send comments and/or bug fixes to hyoon@vilab.com. Any feedback is greatly appreciated.
Please visit HmmLib home page for more information.