hmmus: Posterior decoding with a hidden Markov model

Project Website:

Github Link: None

Description

Hmmus has some C implementations of HMM algorithms with Python bindings, and it is meant to be useful under the following conditions:

The sequence of observations to be analyzed is so long that it does not fit conveniently in RAM.
Likelihoods per hidden state per position have been precalculated.
Numerical stability is important, but is not so important that error bounds on the output are required.
Speed is important.
The number of hidden states is small.
The matrix of probabilities of transitions between hidden states is dense.
Binary data files are acceptable as input and output.
This project would be especially useless in the following cases:

User friendly or pedagogically informative software is desired.
All of the data can fit in RAM and numerical stability is not an issue.
The hidden state transitions are defined by a large sparse graph.
The emission distributions are uncomplicated (e.g. finite or normal).
A variable number of observations are emitted per hidden state.
Silent states other than start and stop states are used.