toy baum welch - Pastebin.com

Abstraction: Python implementation of forward-backward algorithm and Baum-Welch EM for HMMs

Key points:

• Implements forward (alpha) and backward (beta) passes for hidden Markov models in Python
• Follows Jason Eisner's spreadsheet-based teaching tool (JHU, 2002)
• Re-estimates transition, emission, initial, and final probability matrices via EM
• Convergence checked by likelihood ratio threshold (default 4e-20); stops early when ratio met
• Uses JSON files for all probability matrix inputs; observations passed as space-separated string
• Also implements Viterbi best-path decoding over the forward lattice

Connections: Hidden Markov Models · Expectation Maximization

Source: http://pastebin.com/aJG3Ukmn