def _accumulate_sufficient_statistics(self, stats, obs, framelogprob,
posteriors, fwdlattice, bwdlattice,
params):
super(GMMHMM, self)._accumulate_sufficient_statistics(
stats, obs, framelogprob, posteriors, fwdlattice, bwdlattice,
params)
for state, g in enumerate(self.gmms):
lgmm_posteriors = np.log(g.eval(obs)[1] + np.finfo(np.float).eps)
lgmm_posteriors += np.log(posteriors[:, state][:, np.newaxis]
+ np.finfo(np.float).eps)
gmm_posteriors = np.exp(lgmm_posteriors)
tmp_gmm = GMM(g.n_components, cvtype=g.cvtype)
tmp_gmm.n_features = g.n_features
tmp_gmm.covars = _distribute_covar_matrix_to_match_cvtype(
np.eye(g.n_features), g.cvtype, g.n_components)
norm = tmp_gmm._do_mstep(obs, gmm_posteriors, params)
if np.any(np.isnan(tmp_gmm.covars)):
raise ValueError
stats['norm'][state] += norm
if 'm' in params:
stats['means'][state] += tmp_gmm.means * norm[:, np.newaxis]
if 'c' in params:
if tmp_gmm.cvtype == 'tied':
stats['covars'][state] += tmp_gmm._covars * norm.sum()
else:
cvnorm = np.copy(norm)
shape = np.ones(tmp_gmm._covars.ndim)
shape[0] = np.shape(tmp_gmm._covars)[0]
cvnorm.shape = shape
stats['covars'][state] += tmp_gmm._covars * cvnorm
def _init(self, obs, params='stwmc'):
super(GMMHMM, self)._init(obs, params=params)
allobs = np.concatenate(obs, 0)
n_centers = self.n_components * self.n_mix
cluster_centers = cluster.KMeans(
k=n_centers).fit(allobs).cluster_centers_
K = cluster_centers.shape[1]
inds = random.sample(np.arange(0, n_centers), n_centers)
for i in xrange(self.n_components):
self.gmms[i].n_features = K
if 'm' in params:
for i in xrange(self.n_components):
self.gmms[i]._means = cluster_centers[inds[(i * self.n_mix):(
(i + 1) * self.n_mix)], :] + np.random.multivariate_normal(
np.zeros(K),
np.eye(K) * self.var, self.n_mix)
if 'c' in params:
cv = np.cov(obs[0].T)
if not cv.shape:
cv.shape = (1, 1)
for i in xrange(self.n_components):
self.gmms[
i]._covars = _distribute_covar_matrix_to_match_cvtype(
cv, self.cvtype, self.n_mix)
def _init(self, obs, params='stmc', **kwargs):
super(GaussianHMM, self)._init(obs, params=params)
if 'm' in params:
self._means, tmp = sp.cluster.vq.kmeans2(obs[0], self._nstates,
**kwargs)
if 'c' in params:
cv = np.cov(obs[0].T)
if not cv.shape:
cv.shape = (1, 1)
self._covars = _distribute_covar_matrix_to_match_cvtype(
cv, self._cvtype, self._nstates)
def _init(self, obs, params='stmc', **kwargs):
super(GaussianHMM, self)._init(obs, params=params)
if 'm' in params:
self._means, tmp = sp.cluster.vq.kmeans2(obs[0], self._nstates,
**kwargs)
if 'c' in params:
cv = np.cov(obs[0].T)
if not cv.shape:
cv.shape = (1, 1)
self._covars = _distribute_covar_matrix_to_match_cvtype(
cv, self._cvtype, self._nstates)
def __init__(self,
nstates=1,
ndim=1,
cvtype='diag',
startprob=None,
transmat=None,
labels=None,
means=None,
covars=None,
trainer=hmm_trainers.GaussianHMMBaumWelchTrainer()):
"""Create a hidden Markov model with Gaussian emissions.
Initializes parameters such that every state has zero mean and
identity covariance.
Parameters
----------
ndim : int
Dimensionality of the states.
nstates : int
Number of states.
cvtype : string (read-only)
String describing the type of covariance parameters to
use. Must be one of 'spherical', 'tied', 'diag', 'full'.
Defaults to 'diag'.
"""
super(GaussianHMM, self).__init__(nstates, startprob, transmat, labels,
trainer)
self._ndim = ndim
self._cvtype = cvtype
if means is None:
means = np.zeros((nstates, ndim))
self.means = means
if covars is None:
covars = _distribute_covar_matrix_to_match_cvtype(
np.eye(ndim), cvtype, nstates)
self.covars = covars
self.trainer = trainer
def _init(self, obs, params='stmc'):
super(GaussianHMM, self)._init(obs, params=params)
if (hasattr(self, 'n_features')
and self.n_features != obs[0].shape[1]):
raise ValueError('Unexpected number of dimensions, got %s but '
'expected %s' % (obs[0].shape[1],
self.n_features))
self.n_features = obs[0].shape[1]
if 'm' in params:
self._means = cluster.KMeans(
k=self.n_components).fit(obs[0]).cluster_centers_
if 'c' in params:
cv = np.cov(obs[0].T)
if not cv.shape:
cv.shape = (1, 1)
self._covars = _distribute_covar_matrix_to_match_cvtype(
cv, self._cvtype, self.n_components)
def _init(self, obs, params='stwmc'):
super(GMMHMM, self)._init(obs, params=params)
allobs = np.concatenate(obs, 0)
n_centers = self.n_components*self.n_mix
cluster_centers = cluster.KMeans(n_clusters=n_centers).fit(allobs).cluster_centers_
K = cluster_centers.shape[1]
inds = random.sample(np.arange(0,n_centers),n_centers)
for i in xrange(self.n_components):
self.gmms[i].n_features = K
if 'm' in params:
for i in xrange(self.n_components):
self.gmms[i]._means = cluster_centers[inds[(i*self.n_mix):((i+1)*self.n_mix)],:] + np.random.multivariate_normal(np.zeros(K),np.eye(K)*self.var,self.n_mix)
if 'c' in params:
cv = np.cov(obs[0].T)
if not cv.shape:
cv.shape = (1, 1)
for i in xrange(self.n_components):
self.gmms[i]._covars = _distribute_covar_matrix_to_match_cvtype(
cv, self.cvtype, self.n_mix)
def __init__(self, nstates, ndim=1, cvtype='diag', startprob=None,
transmat=None, labels=None, means=None, covars=None,
trainer=hmm_trainers.GaussianHMMBaumWelchTrainer()):
"""Create a hidden Markov model with Gaussian emissions.
Initializes parameters such that every state has zero mean and
identity covariance.
Parameters
----------
nstates : int
Number of states.
ndim : int
Dimensionality of the emissions.
cvtype : string
String describing the type of covariance parameters to
use. Must be one of 'spherical', 'tied', 'diag', 'full'.
Defaults to 'diag'.
"""
super(GaussianHMM, self).__init__(nstates, startprob, transmat, labels)
self._ndim = ndim
self._cvtype = cvtype
if not cvtype in ['spherical', 'tied', 'diag', 'full']:
raise ValueError('bad cvtype')
if means is None:
means = np.zeros((nstates, ndim))
self.means = means
if covars is None:
covars = _distribute_covar_matrix_to_match_cvtype(np.eye(ndim),
cvtype, nstates)
self.covars = covars
self._default_trainer = trainer
