def predict(self, word):
X = self.vec.transform(word)
scores = X.dot(self.coef_.T).toarray()
y = viterbi.decode(scores, self.intercept_trans_, self.intercept_init_, self.intercept_final_)
y = [self.classes_[pred] for pred in y]
return re.sub('_','',''.join(y))
def predict(self, word):
X = self.vec.transform(word)
scores = X.dot(self.coef_.T).toarray()
y = viterbi.decode(scores, self.intercept_trans_, self.intercept_init_,
self.intercept_final_)
y = [self.classes_[pred] for pred in y]
return re.sub('_', '', ''.join(y))
def decode_sentence(s, model):
return decode(s, model)
from corpus import get_data #,get_lexicon
from hmm import get_HMM
from viterbi import decode
import random
# test
#train_set = [ random.choice(sents) for i in range(0,200) ]
train_set = get_data("corpus.xml")[:200]
model = get_HMM(train_set)
s = random.choice(train_set)
#s = sents[4023]
print("phrase: ", s["num"])
s["decoded"] = decode(s, model)
for i in range(0, s["len"]):
if s["tokens"][i] in forget_words:
print("*UNK*\t", s["tokens"][i], "\t", s["tags"][i], "\t",
s["decoded"][i])
else:
print(" \t", s["tokens"][i], "\t", s["tags"][i], "\t",
s["decoded"][i])
for i in xrange(num_states):
trans_probs.append([0]*num_states)
for i in xrange(num_states):
trans_probs[i] = normalise(dict_trans[i],num_states)
return trans_probs
def get_obs_probs(dict_obsInState, num_states, num_obs):
obs_probs = []
for i in xrange(num_states):
obs_probs.append([0]*num_obs)
for i in xrange(num_states):
obs_probs[i] = normalise(dict_obsInState[i],num_obs)
return obs_probs
l = get_data("train.txt")
list_seq = get_seq(l)
a = process_seq(list_seq)
first_seq = list_seq[0]
init_probs = get_initial_probs(a[-1],len(a[1]))
trans_probs = get_trans_probs(a[2],len(a[1]))
obs_probs = get_obs_probs(a[-2],len(a[1]),len(a[0]))
num_states = len(a[1])
num_observs = len(a[0])
obs_seq = [a[0][i[0]] for i in first_seq]
time_steps = len(obs_seq)
l = decode(num_observs,num_states,time_steps,init_probs,trans_probs,obs_probs,obs_seq)