def test_axis(self):
a = np.random.random((10, 5))
out = quantile(a, 0.5, epsilon=1, bounds=(0, 1), axis=0)
self.assertEqual(out.shape, (5, ))
out = quantile(a, 0.5, epsilon=1, bounds=(0, 1), axis=())
self.assertEqual(out.shape, a.shape)
def test_accountant(self):
from diffprivlib.accountant import BudgetAccountant
acc = BudgetAccountant(1.5, 0)
a = np.random.random((1000, 5))
quantile(a, 0.5, epsilon=1, bounds=(0, 1), accountant=acc)
self.assertEqual((1.0, 0), acc.total())
with acc:
with self.assertRaises(BudgetError):
quantile(a, 0.5, epsilon=1, bounds=(0, 1))
def test_accountant_with_axes(self):
from diffprivlib.accountant import BudgetAccountant
acc = BudgetAccountant()
a = np.random.random((1000, 4))
quantile(a, (0.5, 0.75),
epsilon=1,
bounds=(0, 1),
axis=0,
accountant=acc)
# Expecting a different spend on each of the 8 outputs
self.assertEqual((1, 0), acc.total())
self.assertEqual(8, len(acc))
def test_multiple_q(self):
a = np.random.random(1000)
q = [0.1, 0.25, 0.5, 0.75, 0.9]
res = quantile(a, q, epsilon=5, bounds=(0, 1))
self.assertEqual(len(q), len(res))
self.assertTrue(np.isclose(res, q, atol=0.05).all())
def test_normal(self):
a = np.random.normal(size=2000)
res = quantile(a, 0.5, epsilon=3, bounds=(-3, 3))
self.assertAlmostEqual(res, 0, delta=0.1)
def test_simple(self):
a = np.random.random(1000)
for q in [0.1, 0.25, 0.5, 0.75, 0.9]:
res = quantile(a, q, epsilon=5, bounds=(0, 1))
self.assertAlmostEqual(res, q, delta=0.05)
def test_output_type(self):
res = quantile([1, 2, 3], 0.5, bounds=(1, 3))
self.assertTrue(isinstance(res, float))
def test_inf_epsilon(self):
a = [0, 1, 2, 3]
for _ in range(100):
res_dp = quantile(a, 0.25, epsilon=float("inf"), bounds=(0, 3))
self.assertTrue(0 <= res_dp <= 1)
def test_nan(self):
a = np.random.random((5, 5))
a[2, 2] = np.nan
res = quantile(a, 0.5, bounds=(0, 1))
self.assertTrue(np.isnan(res))
def test_missing_bounds(self):
a = np.array([1, 2, 3])
with self.assertWarns(PrivacyLeakWarning):
res = quantile(a, 0.5, epsilon=1, bounds=None)
self.assertIsNotNone(res)
def test_bad_bounds(self):
a = np.array([1, 2, 3])
with self.assertRaises(ValueError):
quantile(a, 0.5, epsilon=1, bounds=(0, -1))
def test_bad_quantile(self):
with self.assertRaises(ValueError):
quantile([1], -0.5, epsilon=1, bounds=(0, 1))
with self.assertRaises(ValueError):
quantile([1], 1.5, epsilon=1, bounds=(0, 1))
def test_no_epsilon(self):
a = np.array([1, 2, 3])
self.assertIsNotNone(quantile(a, 0.5, bounds=(0, 1)))
def test_no_params(self):
a = np.array([1, 2, 3])
with self.assertWarns(PrivacyLeakWarning):
res = quantile(a, 0.5)
self.assertIsNotNone(res)
def test_keepdims(self):
a = np.random.random((10, 5))
out = quantile(a, 0.5, epsilon=1, bounds=(0, 1), keepdims=True)
self.assertEqual(a.ndim, out.ndim)
def test_large_epsilon(self):
a = np.random.random(1000)
res = np.quantile(a, 0.5)
res_dp = quantile(a, 0.5, epsilon=5, bounds=(0, 1))
self.assertAlmostEqual(float(res), float(res_dp), delta=0.01)
def test_array_like(self):
self.assertIsNotNone(quantile([1, 2, 3], 0.5, bounds=(1, 3)))
self.assertIsNotNone(quantile((1, 2, 3), 0.5, bounds=(1, 3)))
def test_clipped_output(self):
a = np.random.random((10, ))
for i in range(100):
self.assertTrue(
0 <= quantile(a, 0.5, epsilon=1e-5, bounds=(0, 1)) <= 1)
