def evaluate_pairwise_metrics(contingency_table: np.ndarray,
N: int,
evaluation: Evaluation,
is_sampled_graph: bool = False):
"""Evaluates the pairwise metrics for goodness of the partitioning. Metrics evaluated:
rand index, adjusted rand index, pairwise recall, pairwise precision.
Parameters
---------
contingency_table : np.ndarray (int)
the contingency table (confusion matrix) comparing the true block assignment to the algorithmically
determined block assignment
N : int
the number of nodes in the confusion matrix
evaluation : Evaluation
stores evaluation results
is_sampled_graph : bool
True if evaluation is for a sampled graph. Default = False
"""
# Compute pair-counting-based metrics
def nchoose2(a):
return misc.comb(a, 2)
num_pairs = nchoose2(N)
colsum = np.sum(contingency_table, axis=0)
rowsum = np.sum(contingency_table, axis=1)
# compute counts of agreements and disagreement (4 types) and the regular rand index
count_in_each_b1 = np.sum(contingency_table, axis=1)
count_in_each_b2 = np.sum(contingency_table, axis=0)
num_same_in_b1 = sum(count_in_each_b1 * (count_in_each_b1 - 1)) / 2
num_same_in_b2 = sum(count_in_each_b2 * (count_in_each_b2 - 1)) / 2
num_agreement_same = 0.5 * sum(
sum(contingency_table * (contingency_table - 1)))
num_agreement_diff = calc_num_agreement_diff(contingency_table, N, colsum,
rowsum)
num_agreement = num_agreement_same + num_agreement_diff
rand_index = num_agreement / num_pairs
adjusted_rand_index = calc_adjusted_rand_index(contingency_table, nchoose2,
colsum, rowsum, num_pairs)
if is_sampled_graph:
evaluation.sampled_graph_rand_index = rand_index
evaluation.sampled_graph_adjusted_rand_index = adjusted_rand_index
evaluation.sampled_graph_pairwise_recall = num_agreement_same / num_same_in_b1
evaluation.sampled_graph_pairwise_precision = num_agreement_same / num_same_in_b2
else:
evaluation.rand_index = rand_index
evaluation.adjusted_rand_index = adjusted_rand_index
evaluation.pairwise_recall = num_agreement_same / num_same_in_b1
evaluation.pairwise_precision = num_agreement_same / num_same_in_b2
print('Rand Index: {}'.format(rand_index))
print('Adjusted Rand Index: {}'.format(adjusted_rand_index))
print('Pairwise Recall: {}'.format(num_agreement_same / (num_same_in_b1)))
print('Pairwise Precision: {}'.format(num_agreement_same /
(num_same_in_b2)))
print('\n')
