causal_metrics: Causal evaluation metrics

In nicolagnecco/causalXtreme: Causal discovery in heavy-tailed models

Causal evaluation metrics

Description

Evaluate the output of the causal inference method called by causal_discovery.

Usage

causal_metrics(simulated_data, estimated_graphs)

Arguments

simulated_data

List returned by simulate_data. The list is made of: dataset — Numeric matrix. Dataset of simulated data with n rows and p columns (note that the hidden variables are not included in this matrix). dag — Square binary matrix. The generated DAG, including both the observed variables and the confounders, if the argument has_confounder = TRUE when calling simulate_data. pos_confounders — Integer vector. Represents the position of confounders (rows and columns) in dag. If the argument has_confounder = FALSE when calling simulate_data, then pos_confounders = integer(0).

estimated_graphs

List returned by causal_discovery. The list is made of: est_g — Square binary matrix (or NA in case of error). The estimated DAG (or CPDAG when the method is pc or pc_rank). est_cpdag — Square binary matrix (or NA in case of error). The estimated CPDAG.

Details

The evaluation is done with respect to the structural intervention distance (see compute_str_int_distance). and the structural Hamming distance (see compute_str_ham_distance).

Value

List. The list is made of:

• sid Numeric — between 0 and 1 — (or NA if est_g is NA). The structural intervention distance between the true DAG dag and the estimated DAG (or CPDAG) est_g. See also compute_str_int_distance.

• shd Numeric — between 0 and 1 — (or NA if est_cpdag is NA). The structural Hamming distance between the true CPDAG (dag_to_cpdag(dag)) and the estimated CPDAG est_cpdag. See also compute_str_ham_distance.

nicolagnecco/causalXtreme documentation built on April 21, 2024, 4:22 a.m.