markovflow.base

Module containing type aliases, attributes and helper functions.

SampleShape[source]

A type that is either a Tuple, a List, or an int.

alias of Union[Tuple, List, int]

Module Contents

ordered()[source]

A bijector to be used when parameterising inducing points so that we ensure they remain ordered. TensorFlow Probability unfortunately got their naming the wrong way around, see https://github.com/tensorflow/probability/issues/765 - we need to construct Invert(Ordered()) to obtain a Parameter that is always ordered!

Note: when using TensorFlow Probability >= 0.12.0 this can be replaced with tfp.bijectors.Ascending.

SampleShape[source]

A type that is either a Tuple, a List, or an int.

APPROX_INF = 10000000000.0[source]

Constant defining an appropriate level of approximate inference.

AUTO_NAMESCOPE = AUTO_NAMESCOPE[source]

Name of environmental variable which if set enables a well structured TensorFlow graph for tensorboard debugging. In rare cases may cause issues, and will clutter the stacktrace so off by default.

auto_namescope_enabled() → bool[source]

Return True if autonamescoping is enabled. See the description of AUTO_NAMESCOPE.