def __init__(self, *args, initialize=True, **kwargs):
super().__init__(*args, dims=self.compute_dims(*args), **kwargs)
# Natural parameters and moments. Note that the size of the
# array self.u or self.phi doesn't actually need to equal
# self.dims+self.plates, it will be broadcasted automatically
# (although this might be extremely rare: it would mean that
# there are identical posterior distributions over
# plates). Anyway, for this reason (and some others), we need
# to explicityly have the dimensionalities dims and plates? Do
# we need dims? Yes, because you need to be able to check what
# plates are missing.
#self.phi = [np.array([]) for i in range(len(self.dims))]
#self.u = [np.array([]) for i in range(len(self.dims))]
#self.phi = [np.array(0.0) for i in range(len(self.dims))]
#self.u = [np.array(0.0) for i in range(len(self.dims))]
axes = len(self.plates) * (1, )
self.phi = [utils.nans(axes + dim) for dim in self.dims]
self.u = [utils.nans(axes + dim) for dim in self.dims]
# self.u = [np.array(0.0) for i in range(len(self.dims))]
# Terms for the lower bound (G for latent and F for observed)
self.g = 0
self.f = 0
# Not observed
self.observed = False
# By default, ignore all plates
self.mask = False
if initialize:
self.initialize_from_prior()
def __init__(self, *args, initialize=True, **kwargs):
super().__init__(*args, initialize=initialize, **kwargs)
if not initialize:
axes = len(self.plates) * (1, )
self.phi = [utils.nans(axes + dim) for dim in self.dims]
# Terms for the lower bound (G for latent and F for observed)
self.g = np.array(0)
self.f = np.array(0)
def __init__(self, *args, initialize=True, **kwargs):
# Terms for the lower bound (G for latent and F for observed)
self.g = np.array(np.nan)
self.f = np.array(np.nan)
super().__init__(*args, initialize=initialize, **kwargs)
if not initialize:
axes = len(self.plates) * (1,)
self.phi = [utils.nans(axes + dim) for dim in self.dims]
def __init__(self, *args, initialize=True, **kwargs):
super().__init__(*args, dims=self.compute_dims(*args), **kwargs)
# Initialize moment array
axes = len(self.plates) * (1, )
self.u = [utils.nans(axes + dim) for dim in self.dims]
# Not observed
self.observed = False
if initialize:
self.initialize_from_prior()
def __init__(self, *args, initialize=True, **kwargs):
super().__init__(*args, dims=self.compute_dims(*args), **kwargs)
# Initialize moment array
axes = len(self.plates) * (1,)
self.u = [utils.nans(axes + dim) for dim in self.dims]
# Not observed
self.observed = False
if initialize:
self.initialize_from_prior()
def __init__(self, *args, initialize=True, **kwargs):
super().__init__(*args,
dims=self.compute_dims(*args),
**kwargs)
# Natural parameters and moments. Note that the size of the
# array self.u or self.phi doesn't actually need to equal
# self.dims+self.plates, it will be broadcasted automatically
# (although this might be extremely rare: it would mean that
# there are identical posterior distributions over
# plates). Anyway, for this reason (and some others), we need
# to explicityly have the dimensionalities dims and plates? Do
# we need dims? Yes, because you need to be able to check what
# plates are missing.
#self.phi = [np.array([]) for i in range(len(self.dims))]
#self.u = [np.array([]) for i in range(len(self.dims))]
#self.phi = [np.array(0.0) for i in range(len(self.dims))]
#self.u = [np.array(0.0) for i in range(len(self.dims))]
axes = len(self.plates)*(1,)
self.phi = [utils.nans(axes+dim) for dim in self.dims]
self.u = [utils.nans(axes+dim) for dim in self.dims]
# self.u = [np.array(0.0) for i in range(len(self.dims))]
# Terms for the lower bound (G for latent and F for observed)
self.g = 0
self.f = 0
# Not observed
self.observed = False
# By default, ignore all plates
self.mask = False
if initialize:
self.initialize_from_prior()
