def resample_states(self, **kwargs):
from autoregressive.messages import resample_arhmm
if len(self.states_list) > 0:
stateseqs = [
np.empty(s.T, dtype='int32') for s in self.states_list
]
params, normalizers = map(
np.array,
zip(*[self._param_matrix(o) for o in self.obs_distns]))
stats, transcounts, loglikes = resample_arhmm(
[s.pi_0.astype(self.dtype) for s in self.states_list],
[s.trans_matrix.astype(self.dtype) for s in self.states_list],
params.astype(self.dtype), normalizers.astype(self.dtype), [
undo_AR_striding(s.data, self.nlags)
for s in self.states_list
], stateseqs, [
np.random.uniform(size=s.T).astype(self.dtype)
for s in self.states_list
], self.alphans)
for s, stateseq, loglike in zip(self.states_list, stateseqs,
loglikes):
s.stateseq = stateseq
s._normalizer = loglike
self._obs_stats = stats
self._transcounts = transcounts
else:
self._obs_stats = None
self._transcounts = []
def _plot_stateseq_data_values(self,
s,
ax,
state_colors,
plot_slice,
update,
data=None):
data = undo_AR_striding(s.data, self.nlags)[plot_slice]
stateseq = np.concatenate(
(np.repeat(s.stateseq[0],
self.nlags), s.stateseq[:-1]))[plot_slice]
colorseq = np.tile(
np.array([state_colors[state] for state in stateseq]),
data.shape[1])
if update and hasattr(s, '_data_lc'):
s._data_lc.set_array(colorseq)
else:
ts = np.arange(data.shape[0])
segments = np.vstack([
AR_striding(np.hstack((ts[:, None], scalarseq[:, None])),
1).reshape(-1, 2, 2) for scalarseq in data.T
])
lc = s._data_lc = LineCollection(segments)
lc.set_array(colorseq)
lc.set_linewidth(0.5)
ax.add_collection(lc)
return s._data_lc
def _generate_obs(self, s):
if s.data is None:
# generating brand new data sequence
data = np.zeros((s.T + self.nlags, self.D))
if hasattr(self, 'prefix'):
data[:self.nlags] = self.prefix
else:
data[:self.nlags] = self.init_emission_distn\
.rvs().reshape(data[:self.nlags].shape)
for idx, state in enumerate(s.stateseq):
data[idx+self.nlags] = \
self.obs_distns[state].rvs(lagged_data=data[idx:idx+self.nlags])
s.data = AR_striding(data, self.nlags)
else:
# filling in missing data
data = undo_AR_striding(s.data, self.nlags)
# TODO should sample from init_emission_distn if there are nans in
# data[:self.nlags]
assert not np.isnan(data[:self.nlags]).any(
), "can't have missing data (nans) in prefix"
nan_idx, = np.where(np.isnan(data[self.nlags:]).any(1))
for idx, state in zip(nan_idx, s.stateseq[nan_idx]):
data[idx+self.nlags] = \
self.obs_distns[state].rvs(lagged_data=data[idx:idx+self.nlags])
return data
def _generate_obs(self,s):
if s.data is None:
# generating brand new data sequence
data = np.zeros((s.T+self.nlags,self.D))
if hasattr(self,'prefix'):
data[:self.nlags] = self.prefix
else:
data[:self.nlags] = self.init_emission_distn\
.rvs().reshape(data[:self.nlags].shape)
for idx, state in enumerate(s.stateseq):
data[idx+self.nlags] = \
self.obs_distns[state].rvs(lagged_data=data[idx:idx+self.nlags])
s.data = AR_striding(data,self.nlags)
else:
# filling in missing data
data = undo_AR_striding(s.data,self.nlags)
# TODO should sample from init_emission_distn if there are nans in
# data[:self.nlags]
assert not np.isnan(data[:self.nlags]).any(), "can't have missing data (nans) in prefix"
nan_idx, = np.where(np.isnan(data[self.nlags:]).any(1))
for idx, state in zip(nan_idx,s.stateseq[nan_idx]):
data[idx+self.nlags] = \
self.obs_distns[state].rvs(lagged_data=data[idx:idx+self.nlags])
return data
def _plot_stateseq_data_values(self,s,ax,state_colors,plot_slice,update,data=None):
data = undo_AR_striding(s.data,self.nlags)[plot_slice]
stateseq = np.concatenate((np.repeat(s.stateseq[0],self.nlags),s.stateseq[:-1]))[plot_slice]
colorseq = np.tile(np.array([state_colors[state] for state in stateseq]),data.shape[1])
if update and hasattr(s,'_data_lc'):
s._data_lc.set_array(colorseq)
else:
ts = np.arange(data.shape[0])
segments = np.vstack(
[AR_striding(np.hstack((ts[:,None], scalarseq[:,None])),1).reshape(-1,2,2)
for scalarseq in data.T])
lc = s._data_lc = LineCollection(segments)
lc.set_array(colorseq)
lc.set_linewidth(0.5)
ax.add_collection(lc)
return s._data_lc
def resample_states(self,**kwargs):
from autoregressive.messages import resample_arhmm
if len(self.states_list) > 0:
stateseqs = [np.empty(s.T,dtype='int32') for s in self.states_list]
params, normalizers = map(np.array,zip(*[self._param_matrix(o) for o in self.obs_distns]))
stats, transcounts, loglikes = resample_arhmm(
[s.pi_0.astype(self.dtype) for s in self.states_list],
[s.trans_matrix.astype(self.dtype) for s in self.states_list],
params.astype(self.dtype), normalizers.astype(self.dtype),
[undo_AR_striding(s.data,self.nlags) for s in self.states_list],
stateseqs,
[np.random.uniform(size=s.T).astype(self.dtype) for s in self.states_list],
self.alphans)
for s, stateseq, loglike in zip(self.states_list,stateseqs,loglikes):
s.stateseq = stateseq
s._normalizer = loglike
self._obs_stats = stats
self._transcounts = transcounts
else:
self._obs_stats = None
self._transcounts = []
def datas(self):
return [undo_AR_striding(s.data, self.nlags) for s in self.states_list]
def _get_joblib_pair(self, s):
return (undo_AR_striding(s.data, self.nlags), s._kwargs)
def datas(self):
return [undo_AR_striding(s.data,self.nlags) for s in self.states_list]
def _get_joblib_pair(self,s):
return (undo_AR_striding(s.data,self.nlags),s._kwargs)
