Structure Learning

group StructureLearning

Bayesian Network Structure Learning (BNSL) algorithms including

  • exhaustive_search

  • simulated_annealing

  • pc

Functions

PDAG exhaustive_search(const DataframeWrapper &df, const ScoreType &score_type, size_t max_parents, double timeout_sec = 300.0)

A function to run the exhaustive search with dynamic programming.

Parameters:

  • df – A DataframeWrapper object

  • max_parents – The maximum number of parents for each node

Returns:

A PDAG object representing the learned structure

PDAG pc_edge_parallel(const DataframeWrapper &df, const CITestType &ci_test_type, size_t max_cond_vars, double timeout_sec)

A function to run the PC algorithm.

Parameters:

  • df – A DataframeWrapper object

  • ci_test_type – The type of conditional independence test to use

  • max_cond_vars – The maximum number of conditional variables to consider

Returns:

A PDAG object

PDAG rai_edge_parallel(const DataframeWrapper &df, const CITestType &ci_test_type, size_t max_cond_vars, double timeout_sec)

A function to run the RAI (Recursive Autonomy Identification) algorithm.

Parameters:

  • df – A DataframeWrapper object

  • ci_test_type – The type of conditional independence test to use

  • max_cond_vars – The maximum number of conditional variables to consider

Returns:

A PDAG object

PDAG simulated_annealing(const DataframeWrapper &df, const ScoreType &score_type, size_t max_parents, size_t max_iters = 100000, double init_temp = 1.0, double cooling_rate = 0.9995, uint64_t seed = 0, size_t num_chains = 0, double timeout_sec = 300.0)

Score-based multi-chain simulated annealing.

Parameters:

  • df – Dataframe containing the data

  • score_type – Type of score to use (e.g., BIC, AIC)

  • max_parents – Maximum number of parents for each node

  • max_iters – Maximum number of iterations for the annealing process

  • init_temp – Initial temperature for the annealing process

  • cooling_rate – Cooling rate for the temperature

  • seed – Random seed for reproducibility

  • num_chains – Number of parallel chains to run (0 for auto-detect)

Returns:

Best DAG found during the annealing process