def test_compute_dp_sum(self):
params = dp_computations.MeanVarParams(
eps=0.5,
delta=1e-10,
min_value=2,
max_value=3,
max_partitions_contributed=1,
max_contributions_per_partition=1,
noise_kind=NoiseKind.LAPLACE)
l0_sensitivity = params.l0_sensitivity()
linf_sensitivity = params.max_contributions_per_partition * max(
params.min_value, params.max_value)
# Laplace Mechanism
l1_sensitivity = dp_computations.compute_l1_sensitivity(
l0_sensitivity, linf_sensitivity)
results = [
dp_computations.compute_dp_sum(sum=10, dp_params=params)
for _ in range(N_ITERATIONS)
]
self._test_laplace_noise(results, 10, params.eps, l1_sensitivity)
# Gaussian Mechanism
params.noise_kind = NoiseKind.GAUSSIAN
l2_sensitivity = dp_computations.compute_l2_sensitivity(
l0_sensitivity, linf_sensitivity)
results = [
dp_computations.compute_dp_sum(sum=10, dp_params=params)
for _ in range(N_ITERATIONS)
]
self._test_gaussian_noise(results, 10, params.eps, params.delta,
l2_sensitivity)
def mean_var_params(self):
return dp_computations.MeanVarParams(
self.eps, self.delta, self.aggregate_params.min_value,
self.aggregate_params.max_value,
self.aggregate_params.max_partitions_contributed,
self.aggregate_params.max_contributions_per_partition,
self.aggregate_params.noise_kind)
def test_l0_sensitivity(self):
params = dp_computations.MeanVarParams(
eps=1,
delta=1e-10,
min_value=2,
max_value=3,
max_partitions_contributed=4,
max_contributions_per_partition=5,
noise_kind=NoiseKind.LAPLACE)
self.assertEqual(params.l0_sensitivity(), 4)
def test_compute_dp_var(self):
params = dp_computations.MeanVarParams(
eps=10,
delta=1e-10,
min_value=1,
max_value=20,
max_partitions_contributed=1,
max_contributions_per_partition=1,
noise_kind=NoiseKind.LAPLACE)
(count_eps,
count_delta), (_, _), (_, _) = dp_computations.equally_split_budget(
params.eps, params.delta, 3)
l0_sensitivity = params.l0_sensitivity()
count_linf_sensitivity = params.max_contributions_per_partition
# Laplace Mechanism
results = [
dp_computations.compute_dp_var(count=100000,
sum=1000000,
sum_squares=20000000,
dp_params=params)
for _ in range(N_ITERATIONS)
]
count_values, sum_values, sum_squares_values, var_values = zip(
*results)
self._test_laplace_noise(
count_values, 100000, count_eps,
dp_computations.compute_l1_sensitivity(l0_sensitivity,
count_linf_sensitivity))
self.assertAlmostEqual(np.mean(sum_values), 1000000, delta=1)
self.assertAlmostEqual(np.mean(sum_squares_values), 20000000, delta=2)
self.assertAlmostEqual(np.mean(var_values), 100, delta=0.1)
# Gaussian Mechanism
params.noise_kind = NoiseKind.GAUSSIAN
results = [
dp_computations.compute_dp_var(count=100000,
sum=1000000,
sum_squares=20000000,
dp_params=params)
for _ in range(N_ITERATIONS)
]
count_values, sum_values, sum_squares_values, var_values = zip(
*results)
self._test_gaussian_noise(
count_values, 100000, count_eps, count_delta,
dp_computations.compute_l2_sensitivity(l0_sensitivity,
count_linf_sensitivity))
self.assertAlmostEqual(np.mean(sum_values), 1000000, delta=5)
self.assertAlmostEqual(np.mean(sum_squares_values), 20000000, delta=5)
self.assertAlmostEqual(np.mean(var_values), 100, delta=0.5)