def posterior_sample(self, data, input=None, mask=None, tag=None):
m = self.state_map
pi0 = self.init_state_distn.initial_state_distn
Ps = self.transitions.transition_matrices(data, input, mask, tag)
log_likes = self.observations.log_likelihoods(data, input, mask, tag)
z_smpl = hmm_sample(replicate(pi0, m), Ps, replicate(log_likes, m))
return self.state_map[z_smpl]
def filter(self, data, input=None, mask=None, tag=None):
m = self.state_map
pi0 = self.init_state_distn.initial_state_distn
Ps = self.transitions.transition_matrices(data, input, mask, tag)
log_likes = self.observations.log_likelihoods(data, input, mask, tag)
pzp1 = hmm_filter(replicate(pi0, m), Ps, replicate(log_likes, m))
return collapse(pzp1, m)
def most_likely_states(self, data, input=None, mask=None, tag=None):
m = self.state_map
pi0 = self.init_state_distn.initial_state_distn
Ps = self.transitions.transition_matrices(data, input, mask, tag)
log_likes = self.observations.log_likelihoods(data, input, mask, tag)
z_star = viterbi(replicate(pi0, m), Ps, replicate(log_likes, m))
return self.state_map[z_star]
def expected_states(self, data, input=None, mask=None, tag=None):
m = self.state_map
log_pi0 = self.init_state_distn.log_initial_state_distn(data, input, mask, tag)
log_Ps = self.transitions.log_transition_matrices(data, input, mask, tag)
log_likes = self.observations.log_likelihoods(data, input, mask, tag)
Ez, Ezzp1, normalizer = hmm_expected_states(replicate(log_pi0, m), log_Ps, replicate(log_likes, m))
# Collapse the expected states
Ez = collapse(Ez, m)
Ezzp1 = collapse(collapse(Ezzp1, m, axis=2), m, axis=1)
return Ez, Ezzp1, normalizer
def log_likelihood(self, datas, inputs=None, masks=None, tags=None):
"""
Compute the log probability of the data under the current
model parameters.
:param datas: single array or list of arrays of data.
:return total log probability of the data.
"""
m = self.state_map
ll = 0
for data, input, mask, tag in zip(datas, inputs, masks, tags):
log_pi0 = self.init_state_distn.log_initial_state_distn(data, input, mask, tag)
log_Ps = self.transitions.log_transition_matrices(data, input, mask, tag)
log_likes = self.observations.log_likelihoods(data, input, mask, tag)
ll += hmm_normalizer(replicate(log_pi0, m), log_Ps, replicate(log_likes, m))
assert np.isfinite(ll)
return ll
