def __init__(self,
a,
ms=None,
Ps=None,
Us=None,
Ss=None,
xs=None,
seed=None):
"""Mixture of Gaussians
Creates a MoG with a valid combination of parameters or an already given
list of Gaussian variables.
Parameters
----------
a : list or np.array, 1d
Mixing coefficients
ms : list, length n_components
Means
Ps : list, length n_components
Precisions
Us : list, length n_components
Precision factors such that U'U = P
Ss : list, length n_components
Covariances
xs : list, length n_components
List of gaussian variables
seed : int or None
If provided, random number generator will be seeded
"""
self.__div__ = self.__truediv__
self.__idiv__ = self.__itruediv__
if ms is not None:
super().__init__(a=np.asarray(a),
ncomp=len(ms),
ndim=np.asarray(ms[0]).ndim,
seed=seed)
if Ps is not None:
self.xs = [
Gaussian(m=m, P=P, seed=self.gen_newseed())
for m, P in zip(ms, Ps)
]
elif Us is not None:
self.xs = [
Gaussian(m=m, U=U, seed=self.gen_newseed())
for m, U in zip(ms, Us)
]
elif Ss is not None:
self.xs = [
Gaussian(m=m, S=S, seed=self.gen_newseed())
for m, S in zip(ms, Ss)
]
else:
raise ValueError('Precision information missing')
elif xs is not None:
super().__init__(a=np.asarray(a),
ncomp=len(xs),
ndim=xs[0].ndim,
seed=seed)
self.xs = xs
else:
raise ValueError('Mean information missing')
def project_to_gaussian(self):
"""Returns a gaussian with the same mean and precision as the mog"""
m, S = self.calc_mean_and_cov()
return Gaussian(m=m, S=S, seed=self.seed)
def __init__(self,
a,
ms=None,
Ps=None,
Us=None,
Ss=None,
xs=None,
seed=None,
beta=0.99):
"""Mixture of Ellipsoidals
Creates a MoE with a valid combination of parameters or an already given
list of Gaussian variables.
Parameters
----------
a : list or np.array, 1d
Mixing coefficients
ms : list, length n_components
Means
Ps : list, length n_components
Precisions
Us : list, length n_components
Precision factors such that U'U = P
Ss : list, length n_components
Covariances
xs : list, length n_components
List of gaussian variables
seed : int or None
If provided, random number generator will be seeded
beta : float
Mass to preserve when sampling
"""
if ms is not None:
super().__init__(a=np.asarray(a),
ncomp=len(ms),
ndim=np.asarray(ms[0]).ndim,
seed=seed)
if Ps is not None:
self.xs = [
Gaussian(m=m, P=P, seed=self.gen_newseed())
for m, P in zip(ms, Ps)
]
elif Us is not None:
self.xs = [
Gaussian(m=m, U=U, seed=self.gen_newseed())
for m, U in zip(ms, Us)
]
elif Ss is not None:
self.xs = [
Gaussian(m=m, S=S, seed=self.gen_newseed())
for m, S in zip(ms, Ss)
]
else:
raise ValueError('Precision information missing')
elif xs is not None:
super().__init__(a=np.asarray(a),
ncomp=len(xs),
ndim=xs[0].ndim,
seed=seed)
self.xs = xs
else:
raise ValueError('Mean information missing')
self.threshold = 2 * gammaincinv(0.5 * self.ndim, beta)