Presently, we assume that the HMMs must always transition to some other node, so they can be run forever. It should be straight-forward to extend this so that we can infer HMMs that run only for a finite amount of time...statistics would be gathered by repeated calls to add_counts_from() modified so that it always begins from the start distribution, rather than continuing from the prior counts.
Presently, we assume that the HMMs must always transition to some other node, so they can be run forever. It should be straight-forward to extend this so that we can infer HMMs that run only for a finite amount of time...statistics would be gathered by repeated calls to add_counts_from() modified so that it always begins from the start distribution, rather than continuing from the prior counts.