graphical_model_learning.unknown_target_igsp

graphical_model_learning.unknown_target_igsp(setting_list: ~typing.List[~typing.Dict], nodes: set, ci_tester: ~conditional_independence.ci_tests.ci_tester.CI_Tester, invariance_tester: ~conditional_independence.invariance_tests.invariance_tester.InvarianceTester, depth: ~typing.Optional[int] = 4, nruns: int = 5, initial_undirected: ~typing.Optional[~typing.Union[str, ~graphical_models.classes.undirected.undirected_graph.UndirectedGraph]] = 'threshold', initial_permutations: ~typing.Optional[~typing.List] = None, verbose: bool = False, use_lowest=True, tup_score=True, no_targets=False) -> (<class 'graphical_models.classes.dags.dag.DAG'>, typing.List[typing.Set[int]])[source]

Use the Unknown Target Interventional Greedy Sparsest Permutation algorithm to estimate a DAG in the I-MEC of the data-generating DAG.

Parameters

  • setting_list – A list of dictionaries that provide meta-information about each non-observational setting.

  • nodes – Nodes in the graph.

  • ci_tester – A conditional independence tester object, which has a method is_ci taking two sets A and B, and a conditioning set C, and returns True/False.

  • invariance_tester – An invariance tester object, which has a method is_invariant taking a node, two settings, and a conditioning set C, and returns True/False.

  • depth – Maximum depth in depth-first search. Use None for infinite search depth.

  • nruns – Number of runs of the algorithm. Each run starts at a random permutation and the sparsest DAG from all runs is returned.

  • initial_undirected – Option to find the starting permutation by using the minimum degree algorithm on an undirected graph that is Markov to the data. You can provide the undirected graph yourself, use the default ‘threshold’ to do simple thresholding on the partial correlation matrix, or select ‘None’ to start at a random permutation.

  • initial_permutations – A list of initial permutations with which to start the algorithm. This option is helpful when there is background knowledge on orders. This option is mutually exclusive with initial_undirected.

  • no_targets – if True, leave out information on known intervention targets.