def evaluate_entropy_metrics(joint_prob: np.ndarray,
evaluation: Evaluation,
is_sampled_graph: bool = False):
"""Evaluates the entropy (information theoretics based) goodness of partition metrics.
Parameters
---------
joint_prob : np.ndarray
the normalized contingency table
evaluation : Evaluation
stores the evaluation metrics
is_sampled_graph : bool = False
True if evaluation is for a sampled_graph. Default = False
"""
# compute the information theoretic metrics
marginal_prob_b2 = np.sum(joint_prob, 0)
marginal_prob_b1 = np.sum(joint_prob, 1)
idx_truth = np.nonzero(marginal_prob_b1)
idx_alg = np.nonzero(marginal_prob_b2)
evaluation = calc_entropy(marginal_prob_b1, marginal_prob_b2, idx_truth,
idx_alg, evaluation, is_sampled_graph)
evaluation = calc_conditional_entropy(joint_prob, marginal_prob_b1,
marginal_prob_b2, idx_truth, idx_alg,
evaluation, is_sampled_graph)
if is_sampled_graph:
if evaluation.sampled_graph_entropy_truth > 0:
fraction_missed_info = (
evaluation.sampled_graph_entropy_truth_given_algorithm /
evaluation.sampled_graph_entropy_truth)
else:
fraction_missed_info = 0
if evaluation.sampled_graph_entropy_algorithm > 0:
fraction_err_info = (
evaluation.sampled_graph_entropy_algorithm_given_truth /
evaluation.sampled_graph_entropy_algorithm)
else:
fraction_err_info = 0
evaluation.sampled_graph_missed_info = fraction_missed_info
evaluation.sampled_graph_erroneous_info = fraction_err_info
else:
if evaluation.entropy_truth > 0:
fraction_missed_info = evaluation.entropy_truth_given_algorithm / evaluation.entropy_truth
else:
fraction_missed_info = 0
if evaluation.entropy_algorithm > 0:
fraction_err_info = evaluation.entropy_algorithm_given_truth / evaluation.entropy_algorithm
else:
fraction_err_info = 0
evaluation.missed_info = fraction_missed_info
evaluation.erroneous_info = fraction_err_info
print('Fraction of missed information: {}'.format(
abs(fraction_missed_info)))
print('Fraction of erroneous information: {}'.format(
abs(fraction_err_info)))
