def _estimate_trans(self):
super(HMMProbBasedEstimator, self)._estimate_trans()
self._trans[:] = 0.0
for i in xrange(self._probs.shape[0]): # tracks
for j in xrange(1, self._probs.shape[1], 1): # frames
prob0 = self._probs[i, j-1, :]
prob1 = self._probs[i, j, :]
condprob = np.matrix(prob0).T*np.matrix(prob1)
self._trans += condprob
self._trans = normalize(self._trans, axis=1, eps=0.0)
def _estimate_startprob(self):
super(HMMTransitionCountEstimator, self)._estimate_startprob()
self.startprob[:] = 0.0
index_of = lambda label: np.where(self.states==label)[0][0]
counts = np.bincount(self._tracks[:, 0].flatten())
for label in np.unique(self._tracks[:, 0]):
self.startprob[index_of(label)] = counts[label]
self._startprob = normalize(self._startprob, eps=0.0)
return self._startprob
def _estimate_emis(self):
"""Estimate the emission matrix by calculating the mean prediction
probabiliy for each class. The result is normalized row wise.
"""
mprobs = np.zeros((self.nstates, self.nstates))
probs = self._probs.reshape((-1, self.nstates))
tracks = self._tracks.flatten()
for i in xrange(self.nstates):
mprobs[i, :] = probs[tracks == i].mean(axis=0)
self._emis = normalize(mprobs, axis=1)
def _set_startprob(self, startprob):
if startprob is None:
startprob = np.tile(1.0 / self.n_components, self.n_components)
else:
startprob = np.asarray(startprob, dtype=np.float)
if not np.alltrue(startprob):
startprob = normalize(startprob, eps=EPS)
if len(startprob) != self.n_components:
raise ValueError('startprob must have length n_components')
if not np.allclose(np.sum(startprob), 1.0):
raise ValueError('startprob must sum to 1.0')
self._log_startprob = np.log(np.asarray(startprob).copy())
def _set_startprob(self, startprob):
if startprob is None:
startprob = np.tile(1.0 / self.n_components, self.n_components)
else:
startprob = np.asarray(startprob, dtype=np.float)
if not np.alltrue(startprob):
startprob = normalize(startprob, eps=EPS)
if len(startprob) != self.n_components:
raise ValueError('startprob must have length n_components')
if not np.allclose(np.sum(startprob), 1.0):
raise ValueError('startprob must sum to 1.0')
self._log_startprob = np.log(np.asarray(startprob).copy())
def _estimate_trans(self):
"""Estimates the transition probaility by counting."""
super(HMMTransitionCountEstimator, self)._estimate_trans()
self._trans[:] = 0.0
index_of = lambda label: np.where(self.states==label)[0][0]
_tracks = self._tracks.flatten()
for i, label in enumerate(_tracks):
for state in self.states:
try:
if (_tracks[i+1] == state):
self._trans[index_of(label), index_of(state)] += 1.0
continue
except IndexError:
pass
self._trans = normalize(self._trans, axis=1, eps=0.0)
return self._trans
def _set_transmat(self, transmat):
if transmat is None:
transmat = np.tile(1.0 / self.n_components,
(self.n_components, self.n_components))
if not np.alltrue(transmat):
transmat = normalize(transmat, axis=1, eps=EPS)
if (np.asarray(transmat).shape \
!= (self.n_components, self.n_components)):
raise ValueError('transmat must have shape '
'(n_components, n_components)')
if not np.all(np.allclose(np.sum(transmat, axis=1), 1.0)):
raise ValueError('Rows of transmat must sum to 1.0')
self._log_transmat = np.log(np.asarray(transmat).copy())
underflow_idx = np.isnan(self._log_transmat)
self._log_transmat[underflow_idx] = -np.inf
def _set_transmat(self, transmat):
if transmat is None:
transmat = np.tile(1.0 / self.n_components,
(self.n_components, self.n_components))
if not np.alltrue(transmat):
transmat = normalize(transmat, axis=1, eps=EPS)
if (np.asarray(transmat).shape \
!= (self.n_components, self.n_components)):
raise ValueError('transmat must have shape '
'(n_components, n_components)')
if not np.all(np.allclose(np.sum(transmat, axis=1), 1.0)):
raise ValueError('Rows of transmat must sum to 1.0')
self._log_transmat = np.log(np.asarray(transmat).copy())
underflow_idx = np.isnan(self._log_transmat)
self._log_transmat[underflow_idx] = -np.inf
def _estimate_emis(self):
"""This emission is an adhoc-assumtiopn!"""
self._emis = normalize(np.eye(self.nstates) + self._emission_noise,
eps=0.0)
def _estimate_emis(self):
"""This emission is an adhoc-assumtiopn!"""
self._emis = normalize(np.eye(self.nstates) + self._emission_noise,
eps=0.0)
def _estimate_startprob(self):
super(HMMProbBasedEstimator, self)._estimate_startprob()
self._startprob[:] = 0.0
self._startprob = self._probs[:, 0, :].sum(axis=0)
self._startprob = normalize(self._startprob, eps=0.0)
def constrain(self, hmmc):
self._trans = normalize(hmmc.trans*self._trans, axis=1, eps=0.0)
self._startprob = normalize(hmmc.start*self._startprob, eps=0.0)
self._emis = normalize(hmmc.emis*self._emis, axis=1, eps=0.0)
def _estimate_emis(self):
self._emis = np.eye(self.nstates) + self._emission_noise
self._emis = normalize(self._emis, axis=1, eps=0.0)
