This software implements Forward-Backward Gibbs sampling for Bayesian segmentation in Hidden Markov Models (HMM). It uses dynamic wavelet compression to drastically improve convergence and memory consumption, making inference possible on large-scale data.
For instance, HaMMLET can be used on a regular laptop for segmentation of genomic data, such as array-CGH or depth-of coverage from whole-genome sequencing (WGS), to find candidates for copy-number variants (CNV). For details, please refer to the doc/ directory.
For implementation details and the theory behind this approach, please refer to my thesis (DOI: 10.7282/t3-4e1k-ph18).