tfsnippet.stochastic

tfsnippet.stochastic.validate_n_samples(value, name)

Validate the n_samples argument.

Parameters:

• value – An int32 value, a int32 tf.Tensor, or None.
• name (str) – Name of the argument (in error message).

Returns:

The validated n_samples argument value.

Return type:

int or tf.Tensor

Raises:

TypeError or ValueError or None – If the value cannot be validated.

class tfsnippet.stochastic.StochasticTensor(distribution, tensor, n_samples=None, group_ndims=0, is_reparameterized=None, log_prob=None)

Bases: tfsnippet.utils.tensor_wrapper.TensorWrapper

Samples or observations of a stochastic variable.

StochasticTensor is a tensor-like object, carrying samples or observations of a random variable, following some distribution of a specific Distribution type.

It mimics the interface of zhusuan.model.StochasticTensor, except that it does not carry a name, and does not add itself to any BayesianNet context automatically. This modified design fits the concept of Module better than the original zhusuan.model.StochasticTensor.

__init__(distribution, tensor, n_samples=None, group_ndims=0, is_reparameterized=None, log_prob=None)

Construct the StochasticTensor.

Parameters:

• distribution (tfsnippet.distributions.Distribution) – The distribution of this StochasticTensor.
• tensor (tf.Tensor or TensorWrapper) – The samples or observations of this StochasticTensor.
• n_samples (tf.Tensor or int) – The number of samples taken in Distribution.sample. If not None, the first dimension of tensor should be the sampling dimension.
• group_ndims (int or tf.Tensor) – The number of dimensions to be considered as events group in samples. (default 0)
• is_reparameterized (bool) – Whether or not the samples are re-parameterized? If not specified, will inherit from tfsnippet.distributions.Distribution.is_reparameterized.
• log_prob (tf.Tensor or None) – Pre-computed log-density of tensor, given group_ndims.

distribution

Get the Distribution of this StochasticTensor.

Returns:The distribution instance. Return type:Distribution

group_ndims

Get the number of dimensions to be considered as events group.

Returns:The configured group_ndims. Return type:int or tf.Tensor

is_continuous

Whether or not this StochasticTensor is continuous?

Returns:Equivalent to self.distribution.is_continuous. Return type:bool

is_reparameterized

Whether or not this StochasticTensor is re-parameterized?

Returns:A boolean indicating whether it is re-parameterized. Return type:bool

log_prob(group_ndims=None)

Compute the log-densities of this StochasticTensor.

Parameters:group_ndims (int or tf.Tensor) – If specified, overriding the configured group_ndims. Returns:The log-densities. Return type:tf.Tensor

n_samples

Get the number of samples taken in Distribution.sample.

Returns:The number of samples. Return type:int or tf.Tensor or None

prob(group_ndims=None)

Compute the densities of this StochasticTensor.

Parameters:group_ndims (int or tf.Tensor) – If specified, overriding the configured group_ndims. Returns:The densities. Return type:tf.Tensor

tensor

Get the samples or observations tensor.

Returns:The tensor specified at construction. Return type:tf.Tensor