def test_sample_bad():
"""Bad clustering should score poorly
"""
clusters = [[1, 1, 0, 0], [0]]
aul1 = aul_score_from_labels(*clusters_to_labels(clusters))
aul2 = aul_score_from_clusters(clusters)
assert_almost_equal(aul1, 0.5, 4)
assert_almost_equal(aul2, 0.5, 4)
auc = RocCurve.from_clusters(clusters).auc_score()
assert_almost_equal(auc, 0.6667, 4)
def test_sample_perverse():
"""Perverese cases are 0.0 < AUL < 0.5
"""
clusters = [[1], [0, 0]]
aul1 = aul_score_from_labels(*clusters_to_labels(clusters))
aul2 = aul_score_from_clusters(clusters)
assert_almost_equal(aul1, 0.1111, 4)
assert_almost_equal(aul2, 0.1111, 4)
auc = RocCurve.from_clusters(clusters).auc_score()
assert_almost_equal(auc, 0.0, 4)
def test_sample_neg_class1():
"""Similar to ``test_sample_perfect`` but have a negative of class 1
"""
clusters = [[1, 1, 1, 1, 1], [1], [0]]
aul1 = aul_score_from_labels(*clusters_to_labels(clusters))
aul2 = aul_score_from_clusters(clusters)
assert_almost_equal(aul1, 0.8690, 4)
assert_almost_equal(aul2, 0.8690, 4)
auc = RocCurve.from_clusters(clusters).auc_score()
assert_almost_equal(auc, 0.9167, 4)
def test_sample_cluster0_c0():
"""Similar to ``test_sample_perfect`` but have a cluster of class 0
"""
clusters = [[1, 1, 1, 1], [0, 0], [0]]
aul1 = aul_score_from_labels(*clusters_to_labels(clusters))
aul2 = aul_score_from_clusters(clusters)
assert_almost_equal(aul1, 0.6667, 4)
assert_almost_equal(aul2, 0.6667, 4)
auc = RocCurve.from_clusters(clusters).auc_score()
assert_almost_equal(auc, 1.0, 4)
def test_sample_cluster0_nh():
"""Same as in ``test_sample_perfect`` but cluster 0 not homogeneous
"""
clusters = [[1, 1, 1, 1, 0], [0], [0]]
aul1 = aul_score_from_labels(*clusters_to_labels(clusters))
aul2 = aul_score_from_clusters(clusters)
assert_almost_equal(aul1, 0.8, 4)
assert_almost_equal(aul2, 0.8, 4)
auc = RocCurve.from_clusters(clusters).auc_score()
assert_almost_equal(auc, 0.8333, 4)
def test_sample_perfect():
"""Perfect clustering
"""
clusters = [[1, 1, 1, 1, 1], [0], [0]]
aul1 = aul_score_from_labels(*clusters_to_labels(clusters))
aul2 = aul_score_from_clusters(clusters)
assert_almost_equal(aul1, 1.0, 4)
assert_almost_equal(aul2, 1.0, 4)
auc = RocCurve.from_clusters(clusters).auc_score()
assert_almost_equal(auc, 1.0, 4)
def add_ranking_metrics(args, clusters, pairs):
"""Add metrics based on ROC and Lift curves
"""
args_metrics = utils.METRICS
if set(utils.ROC_METRICS) & set(args_metrics):
from lsh_hdc.ranking import RocCurve
rc = RocCurve.from_clusters(clusters, is_class_pos=class_is_positive)
if 'roc_auc' in args_metrics:
pairs.append(('roc_auc', rc.auc_score()))
if 'roc_max_info' in args_metrics:
pairs.append(('roc_max_info', rc.max_informedness()))
if set(utils.LIFT_METRICS) & set(args_metrics):
from lsh_hdc.ranking import aul_score_from_clusters as aul_score
clusters_2xc = ([class_is_positive(point) for point in cluster]
for cluster in clusters)
if 'aul_score' in args_metrics:
pairs.append(('aul_score', aul_score(clusters_2xc)))
