def __init__(self,
num_states=None,
alpha=None,
alphav=None,
trans_matrix=None):
self.N = num_states
if trans_matrix is not None:
self._row_distns = [
Multinomial(alpha_0=alpha,
K=self.N,
alphav_0=alphav,
weights=row) for row in trans_matrix
]
elif None not in (alpha, self.N) or (alphav is not None
and alphav.ndim < 2):
self._row_distns = [
Multinomial(alpha_0=alpha, K=self.N, alphav_0=alphav)
for n in range(self.N)
] # sample from prior
# for the case where we pass a matrix for the alpha (used mainly for adding the stickiness)
elif None not in (alpha, self.N) or (alphav is not None
and alphav.ndim == 2):
self._row_distns = [
Multinomial(alpha_0=alpha, K=self.N, alphav_0=alphav[n, :])
for n in range(self.N)
] # sample from prior
def __init__(self,num_states=None,alpha=None,alphav=None,trans_matrix=None):
self.N = num_states
if trans_matrix is not None:
self._row_distns = [Multinomial(alpha_0=alpha,K=self.N,alphav_0=alphav,
weights=row) for row in trans_matrix]
elif None not in (alpha,self.N) or alphav is not None:
self._row_distns = [Multinomial(alpha_0=alpha,K=self.N,alphav_0=alphav)
for n in xrange(self.N)] # sample from prior
def trans_matrix(self, trans_matrix):
N = self.N = trans_matrix.shape[0]
if self.alphav.ndim < 2:
self._row_distns = \
[Multinomial(alpha_0=self.alpha,K=N,alphav_0=self.alphav,weights=row)
for row in trans_matrix]
elif self.alphav.ndim == 2:
self._row_distns = \
[Multinomial(alpha_0=self.alpha,K=N,alphav_0=self.alphav[n,:],weights=row)
for n, row in enumerate(trans_matrix)]
def __init__(self,gamma,alpha,num_states=None,beta=None,trans_matrix=None):
if num_states is None:
assert beta is not None or trans_matrix is not None
self.N = len(beta) if beta is not None else trans_matrix.shape[0]
else:
self.N = num_states
self.alpha = alpha
self.beta_obj = Multinomial(alpha_0=gamma,K=self.N,weights=beta)
super(_WeakLimitHDPHMMTransitionsBase,self).__init__(
num_states=self.N,alpha=alpha,
alphav=alpha*self.beta,trans_matrix=trans_matrix)
def trans_matrix(self, trans_matrix):
N = self.N = trans_matrix.shape[0]
self._row_distns = [
Multinomial(alpha_0=self.alpha,
K=N,
alphav_0=self.alphav,
weights=row) for row in trans_matrix
]
class _WeakLimitHDPHMMTransitionsBase(_HMMTransitionsBase):
def __init__(self,
gamma,
alpha,
num_states=None,
beta=None,
trans_matrix=None):
if num_states is None:
assert beta is not None or trans_matrix is not None
self.N = len(beta) if beta is not None else trans_matrix.shape[0]
else:
self.N = num_states
self.alpha = alpha
self.beta_obj = Multinomial(alpha_0=gamma, K=self.N, weights=beta)
super(_WeakLimitHDPHMMTransitionsBase, self).__init__(
num_states=self.N,
alpha=alpha,
alphav=alpha * self.beta,
trans_matrix=trans_matrix,
)
@property
def beta(self):
return self.beta_obj.weights
@beta.setter
def beta(self, weights):
self.beta_obj.weights = weights
self.alphav = self.alpha * self.beta
@property
def gamma(self):
return self.beta_obj.alpha_0
@gamma.setter
def gamma(self, val):
self.beta_obj.alpha_0 = val
@property
def alpha(self):
return self._alpha
@alpha.setter
def alpha(self, val):
self._alpha = val
def copy_sample(self):
new = super(_WeakLimitHDPHMMTransitionsBase, self).copy_sample()
new.beta_obj = self.beta_obj.copy_sample()
return new
def __init__(self,gamma,alpha,num_states=None,beta=None,trans_matrix=None):
if num_states is None:
assert beta is not None or trans_matrix is not None
self.N = len(beta) if beta is not None else trans_matrix.shape[0]
else:
self.N = num_states
self.alpha = alpha
self.beta_obj = Multinomial(alpha_0=gamma,K=self.N,weights=beta)
super(_WeakLimitHDPHMMTransitionsBase,self).__init__(
num_states=self.N,alpha=alpha,
alphav=alpha*self.beta,trans_matrix=trans_matrix)
class _WeakLimitHDPHMMTransitionsBase(_HMMTransitionsBase):
def __init__(self,gamma,alpha,num_states=None,beta=None,trans_matrix=None):
if num_states is None:
assert beta is not None or trans_matrix is not None
self.N = len(beta) if beta is not None else trans_matrix.shape[0]
else:
self.N = num_states
self.alpha = alpha
self.beta_obj = Multinomial(alpha_0=gamma,K=self.N,weights=beta)
super(_WeakLimitHDPHMMTransitionsBase,self).__init__(
num_states=self.N,alpha=alpha,
alphav=alpha*self.beta,trans_matrix=trans_matrix)
@property
def beta(self):
return self.beta_obj.weights
@beta.setter
def beta(self,weights):
self.beta_obj.weights = weights
self.alphav = self.alpha * self.beta
@property
def gamma(self):
return self.beta_obj.alpha_0
@gamma.setter
def gamma(self,val):
self.beta_obj.alpha_0 = val
@property
def alpha(self):
return self._alpha
@alpha.setter
def alpha(self,val):
self._alpha = val
def copy_sample(self):
new = super(_WeakLimitHDPHMMTransitionsBase,self).copy_sample()
new.beta_obj = self.beta_obj.copy_sample()
return new
