def test_bias_on_unbiased(method, rng):
data = (0, 1, 2, 3)
r = bias(np.mean, data, method=method, random_state=rng)
if method == "balanced":
# bias is exactly zero for linear functions with the balanced bootstrap
assert r == 0
else:
# bias is not exactly zero for ordinary bootstrap
assert r == pytest.approx(0)
def test_bias_on_biased(method, rng):
def biased(x):
return np.var(x, ddof=0)
data = np.arange(100)
bad = biased(data)
correct = np.var(data, ddof=1)
r = bias(biased, data, method=method, size=10000, random_state=rng)
sample_bias = bad - correct
assert r == pytest.approx(sample_bias, rel=0.05)
def test_bias_on_biased_2(method, rng):
def biased(x):
n = len(x)
return (np.sum(x) + 2) / n
data = np.arange(100)
bad = biased(data)
correct = np.mean(data)
r = bias(biased, data, method=method, size=10000, random_state=rng)
sample_bias = bad - correct
assert r == pytest.approx(sample_bias, rel=0.1)