def exp_concrete(self,
name,
temperature,
logits,
n_samples=None,
group_ndims=0,
is_reparameterized=True,
check_numerics=False,
**kwargs):
"""
Add a stochastic node in this :class:`BayesianNet` that follows the
ExpConcrete distribution.
:param name: The name of the stochastic node. Must be unique in a
:class:`BayesianNet`.
See
:class:`~zhusuan.distributions.multivariate.ExpConcrete`
for more information about the other arguments.
:return: A :class:`StochasticTensor` instance.
"""
dist = distributions.ExpConcrete(temperature,
logits,
group_ndims=group_ndims,
is_reparameterized=is_reparameterized,
check_numerics=check_numerics,
**kwargs)
return self.stochastic(name, dist, n_samples=n_samples, **kwargs)
def __init__(self,
temperature,
logits,
is_reparameterized=True,
check_numerics=None):
"""
Construct the :class:`ExpConcrete`.
Args:
temperature: A 0-D `float` Tensor. The temperature of the relaxed
distribution. The temperature should be positive.
logits: An N-D (N >= 1) `float` Tensor of shape
``(..., n_categories)``. Each slice `[i, j,..., k, :]`
represents the un-normalized log probabilities for all
categories. :math:`\\mathrm{logits} \\propto \\log p`
is_reparameterized (bool): Whether or not the gradients can
be propagated through parameters? (default :obj:`True`)
check_numerics (bool): Whether or not to check numerical issues.
Default to ``tfsnippet.settings.check_numerics``.
"""
if check_numerics is None:
check_numerics = settings.check_numerics
super(ExpConcrete, self).__init__(
zd.ExpConcrete(temperature=temperature,
logits=logits,
is_reparameterized=is_reparameterized,
check_numerics=check_numerics))
def __init__(self,
name,
temperature,
logits,
n_samples=None,
group_event_ndims=0,
is_reparameterized=True,
check_numerics=False):
exp_concrete = distributions.ExpConcrete(
temperature,
logits,
group_event_ndims=group_event_ndims,
is_reparameterized=is_reparameterized,
check_numerics=check_numerics)
super(ExpConcrete, self).__init__(name, exp_concrete, n_samples)
def __init__(self,
temperature,
logits,
is_reparameterized=True,
check_numerics=False):
"""
Construct the :class:`ExpConcrete`.
Args:
temperature: A 0-D `float` Tensor. The temperature of the relaxed
distribution. The temperature should be positive.
logits: An N-D (N >= 1) `float` Tensor of shape
``(..., n_categories)``. Each slice `[i, j,..., k, :]`
represents the un-normalized log probabilities for all
categories. :math:`\\mathrm{logits} \\propto \\log p`
"""
super(ExpConcrete, self).__init__(
zd.ExpConcrete(temperature=temperature,
logits=logits,
is_reparameterized=is_reparameterized,
check_numerics=check_numerics))