def init_meanfield_from_sample(self):
self.expected_states = np.hstack(
[(self.stateseq == i).astype("float64")[:, na] for i in range(self.num_states)]
)
from pyhsmm.util.general import count_transitions
self.expected_transcounts = count_transitions(self.stateseq_norep, minlength=self.num_states)
self.expected_durations = expected_durations = np.zeros((self.num_states, self.T))
for state in xrange(self.num_states):
expected_durations[state] += np.bincount(
self.durations_censored[self.stateseq_norep == state], minlength=self.T
)[: self.T]
def init_meanfield_from_sample(self):
self.expected_states = \
np.hstack([(self.stateseq == i).astype('float64')[:,na]
for i in range(self.num_states)])
from pyhsmm.util.general import count_transitions
self.expected_transcounts = \
count_transitions(self.stateseq_norep,minlength=self.num_states)
self.expected_durations = expected_durations = \
np.zeros((self.num_states,self.T))
for state in range(self.num_states):
expected_durations[state] += \
np.bincount(
self.durations_censored[self.stateseq_norep == state],
minlength=self.T)[:self.T]
def init_meanfield_from_sample(self):
# NOTE: only durations is different here; uses Tfull
self.expected_states = \
np.hstack([(self.stateseq == i).astype('float64')[:,na]
for i in range(self.num_states)])
from pyhsmm.util.general import count_transitions
self.expected_transcounts = \
count_transitions(self.stateseq_norep,minlength=self.num_states)
self.expected_durations = expected_durations = \
np.zeros((self.num_states,self.Tfull))
for state in xrange(self.num_states):
expected_durations[state] += \
np.bincount(
self.durations_censored[self.stateseq_norep == state],
minlength=self.Tfull)[:self.Tfull]
def init_meanfield_from_sample(self):
# NOTE: only durations is different here; uses Tfull
self.expected_states = np.hstack([
(self.stateseq == i).astype("float64")[:, na]
for i in range(self.num_states)
])
from pyhsmm.util.general import count_transitions
self.expected_transcounts = count_transitions(
self.stateseq_norep, minlength=self.num_states)
self.expected_durations = expected_durations = np.zeros(
(self.num_states, self.Tfull))
for state in range(self.num_states):
expected_durations[state] += np.bincount(
self.durations_censored[self.stateseq_norep == state],
minlength=self.Tfull,
)[:self.Tfull]
posteriormodel = pyhsmm.models.DiagGaussGMMHSMMPossibleChangepointsSeparateTrans(
alpha=6.,
init_state_concentration=6.,
obs_distns=obs_distns,
dur_distns=dur_distns)
# posteriormodel = pyhsmm.models.DiagGaussHSMMPossibleChangepointsSeparateTrans(
# alpha=6.,init_state_concentration=6.,
# obs_distns=obs_distns,dur_distns=dur_distns)
posteriormodel.add_data(data, changepoints, group_id=0, stateseq=labels)
posteriormodel.init_meanfield_from_sample()
plt.figure()
posteriormodel.plot()
def normalize(A):
return A / A.sum(1)[:, None]
plt.matshow(truemodel.trans_distn.trans_matrix)
from pyhsmm.util.general import count_transitions
plt.matshow(count_transitions(truemodel.states_list[0].stateseq_norep))
plt.matshow(
normalize(
posteriormodel.trans_distns[0].exp_expected_log_trans_matrix)[:N, :N])
plt.show()
components=[pyhsmm.distributions.DiagonalGaussian(**obs_hypparams)
for component in xrange(2)]) for state in xrange(Nmax)]
# obs_distns = [pyhsmm.distributions.DiagonalGaussian(**obs_hypparams) for state in xrange(Nmax)]
dur_distns = [pyhsmm.distributions.PoissonDuration(**dur_hypparams) for state in xrange(Nmax)]
posteriormodel = pyhsmm.models.DiagGaussGMMHSMMPossibleChangepointsSeparateTrans(
alpha=6.,init_state_concentration=6.,
obs_distns=obs_distns,dur_distns=dur_distns)
# posteriormodel = pyhsmm.models.DiagGaussHSMMPossibleChangepointsSeparateTrans(
# alpha=6.,init_state_concentration=6.,
# obs_distns=obs_distns,dur_distns=dur_distns)
posteriormodel.add_data(data,changepoints,group_id=0,stateseq=labels)
posteriormodel.init_meanfield_from_sample()
plt.figure()
posteriormodel.plot()
def normalize(A):
return A / A.sum(1)[:,None]
plt.matshow(truemodel.trans_distn.trans_matrix)
from pyhsmm.util.general import count_transitions
plt.matshow(count_transitions(truemodel.states_list[0].stateseq_norep))
plt.matshow(normalize(posteriormodel.trans_distns[0].exp_expected_log_trans_matrix)[:N,:N])
plt.show()
