def setUp(self):
self.prng = np.random.RandomState(14) # TODO: remove fixed seeding
self.n_unique = 2
self.n_lags = 2
self.startprob = np.array([0.6, 0.4])
self.transmat = np.array([[0.8, 0.2], [0.1, 0.9]])
self.var = np.array([0.1, 0.3])
self.alpha = np.array([0.8, 0.25])
self.shared_alpha = False
self.mu = np.array([[0.7, -2.0, 0.1], [-0.7, 2.0, -0, 1]])
self.precision = np.array([[[599, 394], [800, 834]],
[[375, 353], [342, 353]]])
self.h = ar.ARTHMM(n_unique=self.n_unique,
n_lags=self.n_lags,
random_state=self.prng,
n_features=3,
shared_alpha=self.shared_alpha,
verbose=False)
self.h.precision_ = self.precision
self.h.startprob_ = self.startprob
self.h.transmat_ = self.transmat
self.h.mu_ = self.mu
self.h.var_ = self.var
self.h.alpha_ = self.alpha
def setUp(self):
self.prng = np.random.RandomState(14) # TODO: remove fixed seeding
self.n_unique = 2
self.n_features = 2
self.n_lags = 2
self.startprob = np.array([0.6, 0.4])
self.transmat = np.array([[0.8, 0.2], [0.1, 0.9]])
self.mu = np.array([[5.6, 10.4], [-4.7, 1.2]])
self.alpha = np.array([[0.01, 0.2], [0.003, 0.004]])
self.precision = np.array([[[0.8, 0.1], [0.1, 0.5]],
[[0.9, 0.3], [0.3, 0.8]]])
self.h = ar.ARTHMM(n_unique=self.n_unique,
n_lags=self.n_lags,
random_state=self.prng,
n_features=self.n_features,
shared_alpha=False,
verbose=False,
precision_bounds=np.array([-1e5, 1e5]),
init_params='samt')
self.h.precision_ = self.precision
self.h.startprob_ = self.startprob
self.h.transmat_ = self.transmat
self.h.mu_ = self.mu
self.h.alpha_ = self.alpha
def autohmm(data, target):
'''
vals = np.diff(target['Target'])
data = data[1:]
data = np.column_stack([data, vals])
'''
vals = []
X_train, X_test, Y_train, Y_test = train_test_split(data,
target['Target'],
test_size=0.33,
random_state=42)
states = clustering(target)
model = ar.ARTHMM(n_unique=len(states))
print len(Y_train), X_train.shape
model.fit(X_train) #X_train.values for numpy array as lists of lists
log_proba, state_seq = model.decode(X_test)
print "state seq", len(state_seq), set(state_seq)
path = state_seq
prediction_state = np.argmax(model.transmat_[state_seq[-1], :])
prediction = np.mean(
[state_seq[prediction_state], state_seq[prediction_state - 1]])
#prediction_state = np.argmax(a[path[-1],:])
#prediction = np.argmax(b[prediction_state,:])
vals.append(prediction)
return state_seq, vals
def test_alpha_set_unique():
h = ar.ARTHMM(n_unique=2, n_tied=2, n_lags=3, shared_alpha=False)
h.alpha_ = np.array([[0.1, 0.2, 0.5], [0.2, 0.5, 0.5]])
correct_alpha = np.array([[0.1, 0.2, 0.5], [0.1, 0.2,
0.5], [0.1, 0.2, 0.5],
[0.2, 0.5, 0.5], [0.2, 0.5, 0.5],
[0.2, 0.5, 0.5]])
assert_array_equal(h._alpha_, correct_alpha)
def setUp(self):
self.prng = np.random.RandomState(14) # TODO: remove fixed seeding
self.n_unique = 2
self.n_lags = 1
self.startprob = np.array([0.6, 0.4])
self.transmat = np.array([[0.8, 0.2], [0.1, 0.9]])
self.mu = np.array([2.0, -2.0])
self.var = np.array([0.1, 0.3])
self.alpha = np.array([0.8, 0.25])
self.shared_alpha = False
self.h = ar.ARTHMM(n_unique=self.n_unique,
n_lags=self.n_lags,
random_state=self.prng,
shared_alpha=self.shared_alpha,
verbose=False)
self.h.startprob_ = self.startprob
self.h.transmat_ = self.transmat
self.h.mu_ = self.mu
self.h.var_ = self.var
self.h.alpha_ = self.alpha
def setUp(self):
self.prng = np.random.RandomState(14) # TODO: remove fixed seeding
self.n_unique = 2
self.n_lags = 1
self.startprob = np.array([0.6, 0.4])
self.transmat = np.array([[0.8, 0.2], [0.1, 0.9]])
self.mu = np.array([[-2.0], [12.0]])
self.alpha = np.array([0.08, 0.02])
self.shared_alpha = False
self.precision = np.array([[0.8], [0.9]])
self.h = ar.ARTHMM(n_unique=self.n_unique,
n_lags=self.n_lags,
random_state=self.prng,
shared_alpha=self.shared_alpha,
verbose=False,
init_params='samt',
precision_bounds=np.array([-1e5, 1e5]))
self.h.startprob_ = self.startprob
self.h.precision_ = self.precision
self.h.transmat_ = self.transmat
self.h.mu_ = self.mu
self.h.alpha_ = self.alpha
def test_alpha_getter():
h = ar.ARTHMM(n_unique=2, n_tied=2, n_lags=3, shared_alpha=False)
new_alpha = np.array([[0.1, 0.2, 0.5], [0.2, 0.5, 0.5]])
h.alpha_ = new_alpha
assert_array_equal(h.alpha_, new_alpha)
def test_alpha_unequal_despite_tied():
with assert_raises(ValueError):
h = ar.ARTHMM(n_unique=2, n_lags=1, shared_alpha=True)
h.alpha_ = np.array([[0.1], [0.2]])
def test_alpha_wrong_nlags():
with assert_raises(ValueError):
h = ar.ARTHMM(n_unique=2, n_lags=1, shared_alpha=False)
h.alpha_ = np.array([[0.1, 0.2], [0.2, 0.2]])