Most applications require a handcrafted model topology -- the number of states, the allowed and forbidden transitions -- which are trained with data to produce the final model. To allow experts from the problem domain to effectively interact in the modelling process, we have developed a graphical editor for HMMs called HMMEd which allows to create sophisticated models manually using a graphical user interface. Hierarchical models are supported (e.g. a three state model representating a single codon as one 'super state'), as well as a wide range of HMM extensions and user data associated with the states of the HMM. Graphical editors for discrete emission distributions as well as mixtures of continous pdfs are integrated.
For the exchange of HMMs we propose a XML-based format which is loosely based on GraphML, is hierarchical and also incorporates necessary extensions for proper graphical display.
The GNU (pending permission from the FSF) HMM library (GHMM) is a C-library providing efficient implementations of a comprehensive collection of algorithms for both discrete and continous emission HMMs. Python bindings allow interactive work with HMMs from the Python command line and, at some later stage, tight integration with HMMEd, which is also written in Python using Tkinter.
HMMEd (pronounced Hammered) and the GHMM are licensed under the LGPL and are freely available from http://algorithmics.molgen.mpg.de/ghmm.html respectively http://algorithmics.molgen.mpg.de/ HMMEd.html
