gfn.gym.helpers.bayesian_structure.graph

Functions

sample_erdos_renyi_graph(num_nodes, rng[, p, ...])

Sample an Erdos-Renyi graph.

sample_erdos_renyi_linear_gaussian(num_nodes, rng[, ...])

Sample a linear Gaussian Bayesian network based on an Erdos-Renyi graph.

Module Contents

gfn.gym.helpers.bayesian_structure.graph.sample_erdos_renyi_graph(num_nodes, rng, p=None, num_edges=None, node_names=None)

Sample an Erdos-Renyi graph.

Parameters:

  • num_nodes (int) – Number of nodes in the graph.

  • rng (numpy.random.Generator) – Numpy random number generator.

  • p (Optional[float]) – Probability of creating an edge.

  • num_edges (Optional[int]) – Total number of edges (used to compute p if p is None).

  • node_names (Optional[List[str]]) – Optional list of node names.

Returns:

A PyTorch Geometric Data object representing the sampled graph with

an attribute ‘node_names’ mapping indices to node names.

Return type:

Data

Raises:

ValueError – If both p and num_edges are None.

gfn.gym.helpers.bayesian_structure.graph.sample_erdos_renyi_linear_gaussian(num_nodes, rng, p=None, num_edges=None, node_names=None, loc_edges=0.0, scale_edges=1.0, obs_noise=0.1)

Sample a linear Gaussian Bayesian network based on an Erdos-Renyi graph.

Creates graph structure using torch-geometric and assigns CPD factors for each node. Each CPD factor is constructed by sampling a parameter vector theta (with bias fixed to zero) based on the node’s parent set determined from the graph structure.

Parameters:

  • num_nodes (int) – Number of nodes.

  • rng (numpy.random.Generator) – Random number generator for reproducibility.

  • p (Optional[float]) – Probability of creating an edge.

  • num_edges (Optional[int]) – Total number of edges (used to compute p if p is None).

  • node_names (Optional[List[str]]) – Optional list of node names.

  • loc_edges (float) – Mean value for edge parameters.

  • scale_edges (float) – Standard deviation for edge parameters.

  • obs_noise (float) – Observation noise for each node.

Returns:

A PyTorch Geometric Data object with additional attributes ‘nodes’ and ‘cpds’.

’nodes’ is a list of node names. ‘cpds’ contains a list of LinearGaussianCPD factors for each node.

Return type:

Data