def test_get_private_average_error(self):
nonprivate_points = [[1, 2, 1], [0.4, 0.2, 0.8], [3, 0, 3]]
average_privacy_param = AveragePrivacyParam(epsilon=7,
delta=1e-2,
sensitivity=4.3)
with self.assertRaises(ValueError):
central_privacy_utils.get_private_average(nonprivate_points,
0,
average_privacy_param,
dim=3)
with self.assertRaises(ValueError):
central_privacy_utils.get_private_average(nonprivate_points,
-2,
average_privacy_param,
dim=3)
def test_get_private_average_error(self):
nonprivate_points = [[1, 2, 1], [0.4, 0.2, 0.8], [3, 0, 3]]
clustering_param = test_utils.get_test_clustering_param(
epsilon=10,
delta=1e-2,
frac_sum=0.7,
frac_group_count=0.3,
radius=4.3)
with self.assertRaises(ValueError):
central_privacy_utils.get_private_average(nonprivate_points,
0,
clustering_param,
dim=3)
with self.assertRaises(ValueError):
central_privacy_utils.get_private_average(nonprivate_points,
-2,
clustering_param,
dim=3)
def test_get_private_average_infinite_eps(self):
nonprivate_points = [[1, 2, 1], [0.2, 0.1, 0.8], [3, 0, 3]]
private_count = 3
expected_center = [1.4, 0.7, 1.6]
average_privacy_param = AveragePrivacyParam(epsilon=np.inf,
delta=1e-2,
sensitivity=4.3)
self.assertSequenceAlmostEqual(
central_privacy_utils.get_private_average(nonprivate_points,
private_count,
average_privacy_param,
dim=3), expected_center)
def test_get_private_average(self, nonprivate_points, expected_center,
mock_normal_fn):
private_count = 4
average_privacy_param = AveragePrivacyParam(epsilon=7,
delta=1e-2,
sensitivity=4.3)
result = central_privacy_utils.get_private_average(
nonprivate_points, private_count, average_privacy_param, dim=3)
self.assertSequenceAlmostEqual(result, expected_center)
mock_normal_fn.assert_called_once()
self.assertEqual(mock_normal_fn.call_args[1]['size'], 3)
self.assertAlmostEqual(mock_normal_fn.call_args[1]['scale'],
1.927768,
delta=1e-5)
def test_get_private_average_infinite_eps(self):
nonprivate_points = [[1, 2, 1], [0.2, 0.1, 0.8], [3, 0, 3]]
private_count = 3
expected_center = [1.4, 0.7, 1.6]
clustering_param = test_utils.get_test_clustering_param(
epsilon=np.inf,
delta=1e-2,
frac_sum=0.7,
frac_group_count=0.3,
radius=4.3)
self.assertSequenceAlmostEqual(
central_privacy_utils.get_private_average(nonprivate_points,
private_count,
clustering_param,
dim=3), expected_center)
def get_private_average(self) -> np.ndarray:
"""Returns and saves private average of the points in the node.
Requires that self.private_count >= 1.
"""
# Reuse old results if they've been computed in the past.
if self.private_average is not None:
return self.private_average
privacy_param = self.clustering_param.privacy_param
if privacy_param.privacy_model != clustering_params.PrivacyModel.CENTRAL:
raise NotImplementedError(
f"Currently unsupported privacy model: {privacy_param.privacy_model}"
)
self.private_average = central_privacy_utils.get_private_average(
self.nonprivate_points, self.private_count, self.clustering_param,
self.sim_hash.dim)
return self.private_average
def test_get_private_average(self, nonprivate_points, expected_center,
mock_normal_fn):
private_count = 4
clustering_param = test_utils.get_test_clustering_param(
epsilon=10,
delta=1e-2,
frac_sum=0.7,
frac_group_count=0.3,
radius=4.3)
result = central_privacy_utils.get_private_average(nonprivate_points,
private_count,
clustering_param,
dim=3)
self.assertSequenceAlmostEqual(result, expected_center)
mock_normal_fn.assert_called_once()
self.assertEqual(mock_normal_fn.call_args[1]['size'], 3)
self.assertAlmostEqual(mock_normal_fn.call_args[1]['scale'],
1.927768,
delta=1e-5)