Continuous Partially Hidden Markov Models with uncertain noisy labels

Allows to learn HMM and infere with possibly noisy and uncertain knowledge on hidden states

Standard approaches to HMM learning includes unsupervised learning on the one hand when nothing is known on the hidden structure, and semi-supervised learning on the other hand when some data is accompanied by certain and precise knowledge. We suggest an approach that allows to learn parameters with possibly uncertain and noisy prior on the hidden structure. This code allows to retrieve the results of the paper and include extra examples.