meanstds = [(2., 0.1), (1., 0.1), (1., 0.1)]
hmm = HMM(3, transition_matrix=transition, pdf_mean_and_stds=meanstds)
print(hmm.A[0])
print(hmm.pi)
print(np.random.choice(range(hmm.N), 1, p=hmm.pi))
print(hmm.simulate_state_seq(50))
obs = hmm.simulate_observation_seq(100)
plt.plot(obs, 'ko')
plt.plot(obs, 'k--')
# plt.show()
print(hmm.forward(obs))
transition2 = np.asarray([[0.8, 0.18, 0.02], [0.2, 0.4, 0.4], [0.02, 0.4, 0.58]])
hmm2 = HMM(3, transition_matrix=transition2, pdf_mean_and_stds=meanstds)
# obs = hmm2.simulate_observation_seq(100)
print(hmm2.forward(obs))
print(hmm.backward(obs))
print(hmm2.backward(obs))
print(hmm.backward(obs)
/ hmm2.backward(obs))
print(hmm.forward(obs)
/ hmm2.forward(obs))
print(hmm.backward(obs)
/ hmm2.backward(obs)
/ (hmm.forward(obs)
/ hmm2.forward(obs))) # always around 0.998, 1.078 or 0.947
hmm = HMM()
hmm.train(train)
test(test)
def test(data):
correct_num = 0.0
token_num = 0.0
for seq in data[10:20]:
true = seq.labels
sen = seq.features
ob, T = self.checkOb(seq)
print(sen)
result = self.classify(seq)
print(true)
print(result)
B = hmm.backward(ob, T)
print(sum(B[:, 0] * hmm.P * hmm.E[ob[0]]))
print(hmm.likelihood(seq))
print()
token_num += len(sen)
for i in range(len(sen)):
if true[i] == result[i]:
correct_num += 1
print('\nAccuracy: ' + str(correct_num / token_num))
print(hmm.forward(data[10]) == hmm.backward(data[10])
) # test if the likelihood of Forward algorithm equals to Backforward.
