def test_gei_alpha_more_than_one(self):
np.random.seed(1)
# test bounds
metric = BinaryFairnessMetrics.GeneralizedEntropyIndex()
my_dict = {}
for _ in range(1000):
random = np.random.choice([0, 1], 10)
if str(random) in my_dict:
continue
else:
my_dict[str(random)] = (random[:5], random[5:])
with self.assertWarns(
RuntimeWarning): # division by 0 in certain instances.
# enumerate through all the combinations
alpha = 3
results = []
vals = []
for y_true, y_pred in my_dict.values():
vals.append([y_true, y_pred])
results.append(metric.get_score(y_true, y_pred, alpha=alpha))
assert min(results) == 0
assert max(results) == (np.power(5, alpha - 1) - 1) / (alpha *
(alpha - 1))
def test_gei_normal_invalid(self):
# Metric
metric = BinaryFairnessMetrics.GeneralizedEntropyIndex()
y_true = np.array([0, 1, 1, 0, 1, 1, 1, 0, 1, -1])
y_pred = np.array([0, 0, 0, 0, 0, 1, 1, 1, 1, 1])
with self.assertRaises(ValueError):
metric.get_score(y_true, y_pred)
def test_gei_normal_list(self):
# Metric
metric = BinaryFairnessMetrics.GeneralizedEntropyIndex()
y_true = [0, 1, 1, 0, 1, 1, 1, 0, 1, 0]
y_pred = [0, 0, 0, 0, 0, 1, 1, 1, 1, 1]
score = metric.get_score(y_true, y_pred)
assert isinstance(score, float)
assert np.isclose(score, 0.302, atol=0.01)
def test_gei_normal_df(self):
my_df = pd.DataFrame.from_dict({
'y_true': [0, 1, 1, 0, 1, 1, 1, 0, 1, 0],
'y_pred': [0, 0, 0, 0, 0, 1, 1, 1, 1, 1]
})
# Metric
metric = BinaryFairnessMetrics.GeneralizedEntropyIndex()
# Score
score = metric.get_score(my_df['y_true'], my_df['y_pred'])
assert isinstance(score, float)
assert np.isclose(score, 0.302, atol=0.01)