def test_gaussian_mechanism_respects_random_state(self):
lm0 = noisers.GaussianMechanism(
lambda x: np.array([1., 2., 3.]),
1.,
2.,
.1,
random_state=np.random.RandomState(seed=123))
result0 = lm0(5.)
lm1 = noisers.GaussianMechanism(
lambda x: np.array([1., 2., 3.]),
1.,
2.,
.1,
random_state=np.random.RandomState(seed=123))
result1 = lm1(5.)
self.assertEqual(list(result0), list(result1))
result2 = lm1(5.)
self.assertNotEqual(list(result1), list(result2))
def test_gaussian_mechanism_adds_expected_noise(self):
rs = FakeGaussianRandomState(np.array([2., 4., 6.]))
lm = noisers.GaussianMechanism(lambda x: np.array([1., 2., 3.]),
1.,
2.,
.1,
random_state=rs)
result = lm(5.)
np.testing.assert_array_equal(result, np.array([3., 6., 9.]))
def __init__(self, epsilon, delta, num_queries=1, random_state=None):
"""Instantiates a GaussianEstimateNoiser object.
Args:
epsilon: The differential privacy level.
delta: The differential privacy level.
num_queries: The number of queries for which the noiser is used. Note
that the constructed noiser will be (epsilon, delta)-differentially
private when answering (no more than) num_queries queries.
random_state: Optional instance of numpy.random.RandomState that is used
to seed the random number generator.
"""
# Note that any cardinality estimator will have sensitivity (delta_f) of 1.
self._noiser = noisers.GaussianMechanism(lambda x: x,
1.0,
epsilon,
delta,
num_queries=num_queries,
random_state=random_state)
def test_gaussian_mechanism_works_with_scipy_stats_normal(self): lm = noisers.GaussianMechanism(lambda x: np.array([1., 2., 3.]), 1., 2., .1) result = lm(5.) self.assertLen(result, 3)