def __call__(self):
"""Get the distribution object from the backend"""
if get_backend() == 'pytorch':
import torch.distributions as tod
return tod.cauchy.Cauchy(self.loc, self.scale)
else:
from tensorflow_probability import distributions as tfd
return tfd.Cauchy(self.loc, self.scale)
def _create_dist(self):
scale_diag = tf.nn.softplus(self._scale_variable) if self._softplus_scale \
else self._scale_variable
scale = distribution_util.make_diag_scale(loc=self._loc_variable,
scale_diag=scale_diag,
validate_args=False,
assert_positive=False)
batch_shape, event_shape = distribution_util.shapes_from_loc_and_scale(
self._loc_variable, scale)
return tfp.TransformedDistribution(
distribution=tfp.Cauchy(loc=tf.zeros([], dtype=scale.dtype),
scale=tf.ones([], dtype=scale.dtype)),
bijector=bijectors.AffineLinearOperator(shift=self._loc_variable,
scale=scale),
batch_shape=batch_shape,
event_shape=event_shape,
name="MultivariateCauchyDiag" +
("SoftplusScale" if self._softplus_scale else ""))
def _init_distribution(conditions, **kwargs):
loc, scale = conditions["loc"], conditions["scale"]
return tfd.Cauchy(loc=loc, scale=scale, **kwargs)
def _base_dist(self, alpha: TensorLike, beta: TensorLike, *args, **kwargs):
return tfd.Cauchy(loc=alpha, scale=beta, **kwargs)
def _create_dist(self):
if self._softplus_scale:
return tfd.Cauchy(self._loc_variable, tf.nn.softplus(self._scale_variable))
return tfd.Cauchy(self._loc_variable, self._scale_variable)
