def evaluate_partition(graph: Graph, true_b: np.ndarray,
alg_partition: BlockState, evaluation: Evaluation):
"""Evaluate the output partition against the truth partition and report the correctness metrics.
Compare the partitions using only the nodes that have known truth block assignment.
Parameters
----------
graph : Graph
the graph which was partitioned.
true_b : ndarray (int)
array of truth block assignment for each node. If the truth block is not known for a node, -1 is used
to indicate unknown blocks.
alg_partition : BlockState
the partition result returned by stochastic block partitioning.
evaluation : Evaluation
stores evaluation results
Returns
------
evaluation : Evaluation
the evaluation results, filled in with goodness of partitioning measures
"""
alg_b = alg_partition.get_blocks().get_array()
evaluation.full_graph_description_length = alg_partition.entropy()
evaluation.max_full_graph_description_length = BlockState(
graph,
graph.new_vertex_property("int",
np.arange(graph.num_vertices()))).entropy()
evaluation.full_graph_modularity = modularity(graph,
alg_partition.get_blocks())
if np.unique(true_b).size != 1:
contingency_table, N = create_contingency_table(
true_b, alg_b, evaluation)
evaluation.contingency_table = contingency_table
joint_prob = evaluate_accuracy(contingency_table, evaluation)
evaluate_pairwise_metrics(contingency_table, N, evaluation)
evaluate_entropy_metrics(joint_prob, evaluation)
else:
evaluation.num_blocks_algorithm = max(alg_b) + 1
evaluation.save()
