def test_RxC_metrics():
"""Alternative implementations should coincide for RxC matrices
"""
for _ in xrange(100):
ltrue = np.random.randint(low=0, high=5, size=(20,))
lpred = np.random.randint(low=0, high=5, size=(20,))
cm = ClusteringMetrics.from_labels(ltrue, lpred)
# homogeneity, completeness, V-measure
expected_v = cm.vi_similarity_m3()
expected_hcv = sklearn_hcv(ltrue, lpred)
actual_hcv = cm.entropy_scores()
assert_array_almost_equal(actual_hcv, expected_hcv)
assert_array_almost_equal(actual_hcv[2], expected_v)
# mutual information score
expected_mi = sklearn_mi(ltrue, lpred)
actual_mi = mutual_info_score(ltrue, lpred)
assert_array_almost_equal(actual_mi, expected_mi)
# adjusted mutual information
expected_ami = sklearn_ami(ltrue, lpred)
actual_ami = adjusted_mutual_info_score(ltrue, lpred)
assert_array_almost_equal(actual_ami, expected_ami)
# adjusted rand index
expected_ari = sklearn_ari(ltrue, lpred)
actual_ari = adjusted_rand_score(ltrue, lpred)
assert_array_almost_equal(actual_ari, expected_ari)
def test_adjusted_mutual_info_score():
# Compute the Adjusted Mutual Information and test against known values
labels_a = np.array([1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3])
labels_b = np.array([1, 1, 1, 1, 2, 1, 2, 2, 2, 2, 3, 1, 3, 3, 3, 2, 2])
# Mutual information
mi_1 = mutual_info_score(labels_a, labels_b)
assert_almost_equal(mi_1, 0.41022, 5)
mi_2 = mutual_info_score(labels_b, labels_a)
assert_almost_equal(mi_2, 0.41022, 5)
# Expected mutual information
cm = ClusteringMetrics.from_labels(labels_a, labels_b)
row_totals = np.fromiter(cm.iter_row_totals(), dtype=np.int64)
col_totals = np.fromiter(cm.iter_col_totals(), dtype=np.int64)
emi_1a = emi_cython(row_totals, col_totals) / cm.grand_total
emi_1b = emi_fortran(row_totals, col_totals) / cm.grand_total
assert_almost_equal(emi_1a, 0.15042, 5)
assert_almost_equal(emi_1b, 0.15042, 5)
emi_2a = emi_cython(col_totals, row_totals) / cm.grand_total
emi_2b = emi_fortran(col_totals, row_totals) / cm.grand_total
assert_almost_equal(emi_2a, 0.15042, 5)
assert_almost_equal(emi_2b, 0.15042, 5)
# Adjusted mutual information (1)
ami_1 = adjusted_mutual_info_score(labels_a, labels_b)
assert_almost_equal(ami_1, 0.27502, 5)
ami_2 = adjusted_mutual_info_score(labels_a, labels_b)
assert_almost_equal(ami_2, 0.27502, 5)
# Adjusted mutual information (2)
ami_1 = adjusted_mutual_info_score([1, 1, 2, 2], [2, 2, 3, 3])
assert_equal(ami_1, 1.0)
ami_2 = adjusted_mutual_info_score([2, 2, 3, 3], [1, 1, 2, 2])
assert_equal(ami_2, 1.0)
# Test AMI with a very large array
a110 = np.array([list(labels_a) * 110]).flatten()
b110 = np.array([list(labels_b) * 110]).flatten()
ami = adjusted_mutual_info_score(a110, b110)
assert_almost_equal(ami, 0.37, 2) # not accurate to more than 2 places
