probabilistic_model.learning.region_graph.region_graph

Classes

Region	A region in a region graph.
Partition	A partition in a region graph.
RegionGraph	A region graph is a directed acyclic bipartite graph that represents a (potentially repeated) partition of variables.

Module Contents

class probabilistic_model.learning.region_graph.region_graph.Region(variables: sortedcontainers.SortedSet)

A region in a region graph.

variables: sortedcontainers.SortedSet

__hash__() → int

random_partition(k=2) → typing_extensions.List[typing_extensions.Self]

__repr__() → str

class probabilistic_model.learning.region_graph.region_graph.Partition

A partition in a region graph.

__hash__() → int

class probabilistic_model.learning.region_graph.region_graph.RegionGraph(variables: sortedcontainers.SortedSet, partitions: int = 2, depth: int = 2, repetitions: int = 2, classes: int = 1)

Bases: networkx.DiGraph

A region graph is a directed acyclic bipartite graph that represents a (potentially repeated) partition of variables. Bluntly speaking, a region graph is an ensemble of random Latent-Variable trees.

variables: sortedcontainers.SortedSet

The variables of the region graph.

classes: int

The number of classes. If classes > 1, the circuit is a classifier, if 1 it is a generative model.

partitions: int

The number of partitions to create in each region.

depth: int

The depth of each repetition of the region graph.

repetitions: int

The number of repetitions to create in the region graph. A repetition is another partitioning on the set of variables.

create_random_region_graph()

Create a random region graph according to the classes parameters in-place.

regions() → typing_extensions.Iterable[Region]

Returns:

An iterator over all regions.

partition_nodes() → typing_extensions.Iterable[Partition]

Returns:

An iterator over all partitions.

recursive_split(node: Region)

Recursively split a region into partitions.

Parameters:

node – The node to start inducing from.

property root: Region

Returns:

The root of the region graph.

as_probabilistic_circuit(input_units: int = 5, sum_units: int = 5, key=jax.random.PRNGKey(69)) → typing_extensions.Union[probabilistic_model.probabilistic_circuit.jax.probabilistic_circuit.ProbabilisticCircuit, probabilistic_model.probabilistic_circuit.jax.probabilistic_circuit.ClassificationCircuit]

Convert the region graph to a jax probabilistic circuit. :param input_units: The number of input units to use in each input layer. :param sum_units: The number of sum units to use in each sum layer. :param key: The random key to use for all trainable parameters. :return: The layered circuit in jax.