zipHMM is a library for hidden Markov models that exploits repetitions in strings to greatly speed up the calculations of the log likelihood of a sequence. The library is released under the LGPL license.
The library analyses the input string and finds repetitive patterns and then reduces the string - similar to how compression algorithms compresses strings - by replacing substrings with new symbols. The new symbols correspond to often seen substrings, and we can precompute the probability for an HMM scanning over such strings. The full likelihood can then be computed similar to the traditional forward algorithm, but much faster since the algorithm can skip over often seen strings.
The development version is available from github ziphmm.
The most recent release is:
For documentation, please consult the README file distributed with the source code.