def test_no_range(self):
X = np.random.rand(5, 10)
with self.assertWarns(PrivacyLeakWarning):
_incremental_mean_and_var(X,
epsilon=float("inf"),
bounds=None,
last_mean=0.,
last_variance=None,
last_sample_count=0)
def test_increment_inf_epsilon(self):
X = np.ones((5, 1))
dp_mean, dp_var, dp_count = _incremental_mean_and_var(
X,
epsilon=float("inf"),
bounds=(0, 1),
last_mean=0.,
last_variance=None,
last_sample_count=5)
self.assertAlmostEqual(dp_mean, 0.5, places=5)
self.assertEqual(dp_count, 10)
def test_different_results(self):
X = np.random.rand(10, 5)
mean1, var1, count1 = _incremental_mean_and_var(X,
epsilon=1,
bounds=(0, 1),
last_mean=0.,
last_variance=0.,
last_sample_count=0)
mean2, var2, count2 = _incremental_mean_and_var(X,
epsilon=1,
bounds=(0, 1),
last_mean=0,
last_variance=0,
last_sample_count=0)
self.assertFalse(np.allclose(mean1, mean2, atol=1e-2))
self.assertFalse(np.allclose(var1, var2, atol=1e-2))
self.assertTrue(np.all(count1 == 10),
"Counts should be 10, got %s" % count1)
self.assertTrue(np.all(count2 == 10),
"Counts should be 10, got %s" % count2)
def test_duplicate_dataset(self):
X = np.random.rand(10, 5)
mean1, var1, count1 = _incremental_mean_and_var(X,
epsilon=float("inf"),
bounds=(0, 1),
last_mean=0.,
last_variance=0.,
last_sample_count=0)
mean2, var2, count2 = _incremental_mean_and_var(
X,
epsilon=float("inf"),
bounds=(0, 1),
last_mean=mean1,
last_variance=var1,
last_sample_count=count1)
self.assertTrue(np.allclose(mean1, mean2))
self.assertTrue(np.allclose(var1, var2))
self.assertTrue(np.all(count1 == 10),
"Counts should be 10, got %s" % count1)
self.assertTrue(np.all(count2 == 20),
"Counts should be 20, got %s" % count2)
def test_inf_epsilon(self):
X = np.random.rand(5, 10)
dp_mean, dp_var, dp_count = _incremental_mean_and_var(
X,
epsilon=float("inf"),
bounds=(0, 1),
last_mean=0.,
last_variance=None,
last_sample_count=np.zeros(X.shape[1], dtype=np.int64))
sk_mean, sk_var, sk_count = sk_incremental_mean_and_var(
X,
last_mean=0.,
last_variance=None,
last_sample_count=np.zeros(X.shape[1], dtype=np.int64))
self.assertTrue(np.allclose(dp_mean, sk_mean))
self.assertIsNone(dp_var)
self.assertIsNone(sk_var)
self.assertTrue((dp_count == sk_count).all())
dp_mean, dp_var, dp_count = _incremental_mean_and_var(
X,
epsilon=float("inf"),
bounds=(0, 1),
last_mean=0.,
last_variance=0.,
last_sample_count=np.zeros(X.shape[1], dtype=np.int64))
sk_mean, sk_var, sk_count = sk_incremental_mean_and_var(
X,
last_mean=0.,
last_variance=0.,
last_sample_count=np.zeros(X.shape[1], dtype=np.int64))
self.assertTrue(np.allclose(dp_mean, sk_mean))
self.assertTrue(np.allclose(dp_var, sk_var))
self.assertTrue((dp_count == sk_count).all())