andrey.metrics.to_cpdag

andrey.metrics.to_cpdag(graph)

Canonicalize a DAG to its CPDAG (essential graph); CPDAG/PAG inputs pass through unchanged.

Comparing MEC-identified graphs (PC / GES / …) through their CPDAG lets a DAG and any score-equivalent member of its Markov equivalence class score identically. A digraph (directed, cyclic) has no essential graph and raises.

Return type:

GraphStructure

Examples

>>> import numpy as np
>>> to_cpdag(np.array([[0, 1], [0, 0]])).kind  # a lone edge is unorientable
'cpdag'