def test_sensitivity_norm_list_of_arrays():
class ReshapeToList(Query):
def get(self, data):
return list(np.reshape(data, (20, 30, 40)))
distribution = NormalDistribution(0, 1.5)
sampler = SensitivitySampler()
# L1 norm:
s_max, s_mean = sampler.sample_sensitivity(ReshapeToList(),
L1SensitivityNorm(),
distribution,
n=20 * 30 * 40,
m=285,
gamma=0.33)
assert isinstance(s_max, list)
assert isinstance(s_mean, list)
for i in range(len(s_max)):
assert s_max[i].sum() < 2 * 1.5
assert s_mean[i].sum() < 2 * 1.5
# L2 norm:
s_max, s_mean = sampler.sample_sensitivity(ReshapeToList(),
L2SensitivityNorm(),
distribution,
n=20 * 30 * 40,
m=285,
gamma=0.33)
assert isinstance(s_max, list)
assert isinstance(s_mean, list)
for i in range(len(s_max)):
assert s_max[i].sum() < 2 * 1.5
assert s_mean[i].sum() < 2 * 1.5
def test_sample_sensitivity_gamma():
distribution = NormalDistribution(0, 1)
sampler = SensitivitySampler()
_, mean = sampler.sample_sensitivity(Mean(),
L1SensitivityNorm(),
distribution,
n=100,
gamma=0.33)
assert np.abs(mean - 0) < 0.5
def test_concatenate_overload_lists_of_arrays():
list_a = [
np.random.rand(30, 40),
np.random.rand(20, 50),
np.random.rand(60, 80)
]
list_b = [
np.random.rand(1, 40),
np.random.rand(1, 50),
np.random.rand(1, 80)
]
sampler = SensitivitySampler()
concatenated_list = sampler._concatenate(list_a, list_b)
for array_i, array_j, array_k in zip(list_a, list_b, concatenated_list):
assert array_k.shape[0] == array_i.shape[0] + array_j.shape[0]
assert array_k.shape[1] == array_i.shape[1] == array_j.shape[1]
def test_concatenate_overload_dictionary_of_arrays():
dict_a = {
1: np.random.rand(30, 40),
2: np.random.rand(20, 50),
3: np.random.rand(60, 80)
}
dict_b = {
3: np.random.rand(1, 40),
4: np.random.rand(1, 50),
5: np.random.rand(1, 80)
}
sampler = SensitivitySampler()
concatenated_dict = sampler._concatenate(dict_a, dict_b)
for i, j, k in zip(dict_a, dict_b, concatenated_dict):
assert concatenated_dict[k].shape[0] == \
dict_a[i].shape[0] + dict_b[j].shape[0]
assert concatenated_dict[k].shape[1] == \
dict_a[i].shape[1] == dict_b[j].shape[1]