def viterbi_smoothing(self, observation, states, emission_matrix,
transition_matrix):
#, observation, states, emission_matrix, transition_matrix):
# initialization
#print observation
#print states
#observation = ['Janet', 'will', 'back', 'the', 'billlklklk', '.']
#states = ['NNP', 'MD', 'VB', 'JJ', 'NN', 'RB' , 'DT']
##states = ['NNP', 'NNS', 'VBZ', '.']
#observation = ['To', 'be', 'sure', ',', 'Kao', 'would', "n't", 'have', 'an', 'easy', 'time', 'taking', 'U.S.', 'market',
# 'share', 'away', 'from', 'the', 'mighty', 'P&G', ',', 'which', 'has', 'about', '23', '%', 'of', 'the',
# 'market', '.']
#states = ['MD', 'VB', 'VBG', 'JJ', 'NN', 'RBR', ',', '.', 'VBN', 'TO', 'VBP', 'WDT', 'RB', 'IN', 'RP', 'DT', 'CD', 'NNS',
#'VBZ', 'NNP']
#observation =['In', 'addition', 'to', '$', '33', 'million', 'compensatory', 'damages', ',', 'the', 'suit', 'seeks', '$',
# '100', 'million', 'in', 'punitive', 'damages', '.']
#states = ['VB', 'RP', '$', 'NN', 'FW', 'RBR', 'DT', 'CD', 'TO', 'RB', 'IN', 'VBZ', 'JJ', '.', 'NNS', ',', 'NNP']
#
#c_1 = Corpus("dataset\\train.col")
#c_2 = Corpus("dataset\\train.col")
#
#emission_matrix = c_1.dictionary_maker(c_1)
#transition_matrix = c_2.bigram_preprocess(c_2)
#print transition_matrix.keys()
#for x in observation:
# print x, emission_matrix[x]
#for wordem in observation:
# print wordem, emission_matrix[wordem]
#for tagtm in states:
# print tagtm, transition_matrix[tagtm]
viterbi = numpy.zeros((len(observation), len(states)))
for ini_state in states:
if observation[0] in emission_matrix:
if ini_state in emission_matrix[observation[0]].keys():
viterbi[0][states.index(ini_state)] = \
emission_matrix[observation[0]][ini_state] * transition_matrix[ini_state]['<S>']
else:
viterbi[0][states.index(
ini_state)] = transition_matrix[ini_state]['<S>']
#print viterbi
#a = [1,2,3,4]
#b = [2,2,2,2]
#c = numpy.multiply(a,b)
#print c
# H@des was here
# He turned coffee to viterbi HMM decoder
#start = time.time()
token = Token()
for word in observation:
if word in emission_matrix:
if observation.index(word) == 0:
pass
else:
# print word
for prev_state in states:
for nex_state in states:
feature_factor = token.perceptron_HMM_feature(
word, nex_state)
if nex_state in emission_matrix[word].keys():
#print word, prev_state, nex_state, max(viterbi[observation.index(word) - 1]), transition_matrix[nex_state][prev_state] , emission_matrix[word][nex_state]
viterbi[observation.index(word)][states.index(nex_state)] = (transition_matrix[prev_state][nex_state] * \
emission_matrix[word][nex_state]) + \
feature_factor
else:
viterbi[observation.index(word)][states.index(
prev_state)] = 0.0
else:
if observation.index(word) == 0:
pass
else:
# print word
for prev_state in states:
for nex_state in states:
feature_factor = token.perceptron_HMM_feature(
word, nex_state)
viterbi[observation.index(word)][states.index(nex_state)] = transition_matrix[prev_state][nex_state] + \
feature_factor
most_probable_tag_sequence = []
for word in observation:
if word != ".":
possible_tags = []
for state in states:
possible_tags.append(
viterbi[observation.index(word)][states.index(state)])
index, value = max(enumerate(possible_tags),
key=operator.itemgetter(1))
most_probable_tag_sequence.append(states[index])
else:
most_probable_tag_sequence.append(".")
#print viterbi
#print most_probable_tag_sequence
#print most_probable_tag_sequence
#print observation
return most_probable_tag_sequence, observation
