def fit_coin():
"""
example from https://github.com/lazyprogrammer/machine_learning_examples/blob/master/hmm_class/hmmd.py
:return:
"""
X = []
for line in open('./data/coin_data.txt'):
# 1 for H, 0 for T
x = [1 if e == 'H' else 0 for e in line.rstrip()]
X.append(x)
X = np.array(X)
print(X.shape)
hmm = HMM(2,2)
hmm.fit(X,max_iter=50)
L = hmm.log_likelihood(X).sum()
print("LL with fitted params:", L)
# try true values
hmm.pi = np.array([0.5, 0.5])
hmm.A = np.array([[0.1, 0.9], [0.8, 0.2]])
hmm.B = np.array([[0.6, 0.4], [0.3, 0.7]])
L = hmm.log_likelihood(X).sum()
print("LL with true params:", L)
# try viterbi
print("Best state sequence for:", X[0])
print(hmm.viterbi(X[0]))
print("MAX ITERATIONS DONE! (", max_iteration, ")")
if iteration % 10 == 0:
print("ITERATION", iteration)
# alpha pass
alpha, c = alpha_pass(model, obs)
# beta pass
beta = beta_pass(model, obs, c)
# gamma-pass
start_probs, trans, emis, gamma, di_gamma = \
Forward_Backward(model.get_N(), alpha, beta, obs.obs, sequence_syms, model.A, model.B)
# update model
model.A = trans
model.B = emis
model.pi = start_probs
# compute lob probs
new_log_prob = compute_logprob(obs, c)
# compare
log_prob_history = np.append(log_prob_history, new_log_prob)
if new_log_prob > old_log_prob:
old_log_prob = new_log_prob
continue
else:
print("STOPPED AT ITERATION:", iteration + 1)
break
"""view some results:"""
# nice pandas display
