def sample_prior(self,
n: int = 1,
seed: int = 1,
two_stage: bool = True) -> tf.Tensor:
if two_stage:
z = atleast_2d(self.latents2.sample(sample_shape=n, seed=seed))
z = self.decoder2(z, training=False)
qz = self.observation2(z, trainable=False)
return tf.convert_to_tensor(qz)
else:
return atleast_2d(self.latents.sample(sample_shape=n, seed=seed))
def sample_prior(self, sample_shape=(), seed=1):
r""" Sampling from prior distribution """
samples = []
for latent in self.latent_layers:
s = bk.atleast_2d(
latent.sample(sample_shape=sample_shape, seed=seed))
samples.append(s)
return samples[0] if len(samples) == 1 else tuple(samples)
def sample_data(self, sample_shape=(), seed=1):
r""" Sample from p(X) given that the prior of X is known, this could be
wrong since `RandomVariable` often has a default prior. """
samples = []
for output in self.output_layers:
s = bk.atleast_2d(
output.sample(sample_shape=sample_shape, seed=seed))
samples.append(s)
return samples[0] if len(samples) == 1 else tuple(samples)
def sample_prior(self,
sample_shape: Union[int, List[int]] = (),
seed: int = 1) -> Union[Tensor, List[Tensor]]:
r""" Sampling from prior distribution """
samples = []
for latent in self.latents:
s = bk.atleast_2d(latent.sample(sample_shape=sample_shape, seed=seed))
samples.append(s)
return samples[0] if len(samples) == 1 else tuple(samples)
def sample_observation_prior(self, n: int = 1, seed: int = 1) -> Tensor:
"""Sample from observation prior, i.e. p(X), given that the prior of X
is known.
Notes
-----
This could be wrong since `RVmeta` often has a default prior.
"""
return bk.atleast_2d(self.observation.sample(sample_shape=n,
seed=seed))
def sample_data(self,
sample_shape: Union[int, List[int]] = (),
seed: int = 1) -> Union[Tensor, List[Tensor]]:
r""" Sample from p(X) given that the prior of X is known, this could be
wrong since `RandomVariable` often has a default prior. """
samples = []
for output in self.observation:
s = bk.atleast_2d(output.sample(sample_shape=sample_shape, seed=seed))
samples.append(s)
return samples[0] if len(samples) == 1 else tuple(samples)
def sample_labels(self,
sample_shape: List[int] = (),
seed: int = 1) -> tf.Tensor:
r""" Sample labels from prior of the labels distribution. """
return bk.atleast_2d(
self.labels.prior.sample(sample_shape=sample_shape, seed=seed))
def sample_prior(self, n: int = 1, seed: int = 1, **kwargs) -> Tensor:
"""Sampling from prior distribution"""
return bk.atleast_2d(self.latents.sample(sample_shape=n, seed=seed))
def sample_labels(self, sample_shape=(), seed=1):
return bk.atleast_2d(
self.labels.prior.sample(sample_shape=sample_shape, seed=seed))
