Viterbi.py

# -*- coding: utf-8 -*-
# @Author: WuLC
# @Date:   2017-04-02 08:52:24
# @Last Modified by:   WuLC
# @Last Modified time: 2017-04-02 09:50:31

###########################################################################################################
# Viterbi Algorithm for HMM
# dynamic programming, time complexity O(mn^2), m is the length of sequence of observation, n is the number of hidden states
# more details can be obtained from: https://tinyurl.com/y8xt2lcf
##########################################################################################################


# five elements for HMM
states = ('Healthy', 'Fever')
 
observations = ('normal', 'cold', 'dizzy')
 
start_probability = {'Healthy': 0.6, 'Fever': 0.4}
 
transition_probability = {
   'Healthy' : {'Healthy': 0.7, 'Fever': 0.3},
   'Fever' :   {'Healthy': 0.4, 'Fever': 0.6},
   }
 
emission_probability = {
   'Healthy' : {'normal': 0.5, 'cold': 0.4, 'dizzy': 0.1},
   'Fever'   : {'normal': 0.1, 'cold': 0.3, 'dizzy': 0.6},
   }



def Viterbit(obs, states, s_pro, t_pro, e_pro):
	path = { s:[] for s in states} # init path: path[s] represents the path ends with s
	curr_pro = {}
	for s in states:
		curr_pro[s] = s_pro[s]*e_pro[s][obs[0]]
	for i in xrange(1, len(obs)):
		last_pro = curr_pro
		curr_pro = {}
		for curr_state in states:
			max_pro, last_sta = max(((last_pro[last_state]*t_pro[last_state][curr_state]*e_pro[curr_state][obs[i]], last_state) 
				                       for last_state in states))
			curr_pro[curr_state] = max_pro
			path[curr_state].append(last_sta)

	# find the final largest probability
	max_pro = -1
	max_path = None
	for s in states:
		path[s].append(s)
		if curr_pro[s] > max_pro:
			max_path = path[s]
			max_pro = curr_pro[s]
		# print '%s: %s'%(curr_pro[s], path[s]) # different path and their probability
	return max_path


if __name__ == '__main__':
	obs = ['normal', 'cold', 'dizzy']
	print Viterbit(obs, states, start_probability, transition_probability, emission_probability)