def resample_gaussian_states(self):
self._aBl = None # clear any caching
self._gaussian_normalizer, self.gaussian_states = \
filter_and_sample(
self.mu_init, self.sigma_init,
self.As, self.BBTs, self.Cs, self.DDTs,
self.data)
def check_filter_and_sample(A, B, C, d, mu_init, sigma_init, data):
rngstate = np.random.get_state()
ll1, sample1 = filter_and_sample(mu_init, sigma_init, A, B.dot(B.T), C, np.diag(d ** 2), data)
np.random.set_state(rngstate)
ll2, sample2 = filter_and_sample_diagonal(mu_init, sigma_init, A, B.dot(B.T), C, d ** 2, data)
assert np.isclose(ll1, ll2)
assert np.allclose(sample1, sample2)
def check_filter_and_sample(A, B, C, d, mu_init, sigma_init, data):
rngstate = np.random.get_state()
ll1, sample1 = filter_and_sample(mu_init, sigma_init, A, B.dot(B.T), C,
np.diag(d**2), data)
np.random.set_state(rngstate)
ll2, sample2 = filter_and_sample_diagonal(mu_init, sigma_init, A,
B.dot(B.T), C, d**2, data)
assert np.isclose(ll1, ll2)
assert np.allclose(sample1, sample2)
def dense_sample_states(sigmasq_states, sigmasq_obs, mu_init, sigmasq_init, data):
T, n = data.shape
# construct corresponding dense model
A = np.eye(n)
sigma_states = np.diag(sigmasq_states)
C = np.eye(n)
sigma_obs = np.diag(sigmasq_obs)
sigma_init = np.diag(sigmasq_init)
return filter_and_sample(
mu_init, sigma_init, A, sigma_states, C, sigma_obs, data)
def dense_sample_states(sigmasq_states, sigmasq_obs, mu_init, sigmasq_init, data):
T, n = data.shape
# construct corresponding dense model
A = np.eye(n)
sigma_states = np.diag(sigmasq_states)
C = np.eye(n)
sigma_obs = np.diag(sigmasq_obs)
sigma_init = np.diag(sigmasq_init)
return filter_and_sample(
mu_init, sigma_init, A, sigma_states, C, sigma_obs, data)
def check_filter_and_sample(A, B, sigma_states, C, D, sigma_obs, mu_init,
sigma_init, inputs, data):
rngstate = np.random.get_state()
ll1, sample1 = filter_and_sample(mu_init, sigma_init, A, B, sigma_states,
C, D, sigma_obs, inputs, inputs)
np.random.set_state(rngstate)
ll2, sample2 = filter_and_sample_diagonal(mu_init, sigma_init, A, B,
sigma_states, C, D,
np.diag(sigma_obs), inputs, data)
assert np.isclose(ll1, ll2)
assert np.allclose(sample1, sample2)
def dense_sample_states(sigmasq_states, sigmasq_obs, mu_init, sigmasq_init,
data):
T, n = data.shape
inputs = np.zeros((T, 0))
# construct corresponding dense model
A = np.eye(n)
B = np.zeros((n, 0))
sigma_states = np.diag(sigmasq_states)
C = np.eye(n)
D = np.zeros((n, 0))
sigma_obs = np.diag(sigmasq_obs)
sigma_init = np.diag(sigmasq_init)
return filter_and_sample(mu_init, sigma_init, A, B, sigma_states, C, D,
sigma_obs, inputs, data)
