def test_list_app_ids(self):
ranking_A = 'acbd'
ranking_B = 'aefg'
references = convert_ranking.list_app_ids(ranking_A,
ranking_B,
verbose=True)
self.assertEqual(len(references), 7)
def test_convert_ranking_to_vector_of_scores(self):
ranking = ["a", "c", "b", "d"]
references = convert_ranking.list_app_ids(ranking, [])
scores = convert_ranking.convert_ranking_to_vector_of_scores(
ranking, references)
self.assertEqual(len(scores), len(ranking))
self.assertGreaterEqual(min(scores), 0)
self.assertLessEqual(max(scores), len(ranking))
def test_convert_ranking_for_scipy(self):
ranking = ["a", "c", "b", "d"]
references = convert_ranking.list_app_ids(ranking, [])
ranks = convert_ranking.convert_ranking_for_scipy(ranking,
references,
reverse_order=False)
self.assertEqual(len(ranks), len(ranking))
self.assertGreaterEqual(min(ranks), 1)
self.assertLessEqual(max(ranks), len(ranking) + 1)
def compute_tau(ranking_A, ranking_B):
# Arrays of ranks
app_ids = list_app_ids(ranking_A, ranking_B)
x = convert_ranking_to_vector_of_ranks(ranking_A, app_ids=app_ids)
y = convert_ranking_to_vector_of_ranks(ranking_B, app_ids=app_ids)
# NB: we would get the same tau with arrays of scores.
tau, p_value = stats.kendalltau(x, y)
print('Kendall rank-order correlation coefficient: {:.4f}'.format(tau))
print('p-value to test for non-correlation: {:.4f}'.format(p_value))
return tau, p_value
def compute_rho(ranking_A, ranking_B):
# Arrays of ranks
app_ids = list_app_ids(ranking_A, ranking_B, verbose=True)
x = convert_ranking_to_vector_of_ranks(ranking_A, app_ids=app_ids)
y = convert_ranking_to_vector_of_ranks(ranking_B, app_ids=app_ids)
# NB: we would get the same rho with arrays of scores.
rho, p_value = stats.spearmanr(x, y)
print('Spearman rank-order correlation coefficient: {:.4f}'.format(rho))
print('p-value to test for non-correlation: {:.4f}'.format(p_value))
return rho, p_value
def compute_weighted_tau(ranking_A, ranking_B):
# Arrays of scores
app_ids = list_app_ids(ranking_A, ranking_B)
x = convert_ranking_to_vector_of_scores(ranking_A, app_ids=app_ids)
y = convert_ranking_to_vector_of_scores(ranking_B, app_ids=app_ids)
# NB: it is important NOT to feed arrays of ranks for the weighted tau!
#
# > Note that if you are computing the weighted on arrays of ranks, rather than of scores (i.e., a larger value
# > implies a lower rank) you must negate the ranks, so that elements of higher rank are associated with a larger
# > value.
#
# Reference: http://scipy.github.io/devdocs/generated/scipy.stats.weightedtau.html#scipy.stats.weightedtau
weighted_tau, p_value = stats.weightedtau(x, y)
print('Weighted Kendall rank-order correlation coefficient: {:.4f}'.format(
weighted_tau))
print('p-value to test for non-correlation: {:.4f}'.format(p_value))
return weighted_tau, p_value