def test_simple_gmm_data():
test_data = data_gen_helper(100, 2)
test_labels = np.random.choice([0, 1], size=100, p=[1. / 2, 1. / 2])
test_ron_gauss_instance = ron_gauss.RONGauss(algorithm="gmm")
dp_data = test_ron_gauss_instance.generate_dpdata(test_data, 1,
test_labels)
assert normality_test_helper(dp_data)
def test_normalize_sample_wise():
number_of_trials = 30
for i in range(number_of_trials):
test_data = np.random.normal(size=(100, 10))
benchmark = preprocessing.normalize(test_data)
new_implementation = ron_gauss.RONGauss()._normalize_sample_wise(
test_data)
assert (np.abs(new_implementation - benchmark) < 1e-15).all()
def test_data_preprocessing():
test_data = np.load('./tests/test_data/test_data_preprocessing.npz')
input = test_data['input']
output = test_data['output']
prng = np.random.RandomState(seed=7)
x_bar, mu_dp = ron_gauss.RONGauss()._data_preprocessing(input,
epsilon_mean=1.0,
prng=prng)
assert check_element_diff(output[0], x_bar) and check_element_diff(
output[1], mu_dp)
def test_simple_supervised_data():
test_data = data_gen_helper(1000, 3)
test_labels = np.random.choice([0, 1], size=1000, p=[1. / 2, 1. / 2])
test_ron_gauss_instance = ron_gauss.RONGauss(algorithm="supervised")
# Implicit testing of successful execution with "supervised" option.
dp_data = test_ron_gauss_instance.generate_dpdata(X=test_data,
dimension=2,
y=test_labels,
maxY=np.array(1))
assert normality_test_helper(dp_data)
def test_dimensionality_agnostic_execution():
for i in np.arange(2, 101):
rand_num_pts = np.random.randint(100) + 100
# Generate a random number of normally distributed data points in
# the range of 101-200 in i dimensions.
test_data = data_gen_helper(rand_num_pts, i)
# Simulate DP-safe lower dimensionality data.
test_ron_gauss_instance = ron_gauss.RONGauss(algorithm="unsupervised")
dims_reduced = int(np.ceil((i - 1) * (rand_num_pts / 200)))
dp_data = test_ron_gauss_instance.generate_dpdata(
test_data, dims_reduced)
# Test dimensions are as expected.
assert dp_data[0].shape[0] == rand_num_pts
assert dp_data[0].shape[1] == dims_reduced
# Test that the reconstruct parameter is working regardless of projection dimension
dp_data = test_ron_gauss_instance.generate_dpdata(test_data,
dims_reduced,
reconstruct=True)
# Test dimensions are as expected.
assert dp_data[0].shape[0] == rand_num_pts
assert dp_data[0].shape[1] == i
def test_simple_unsupervised_data():
test_data = data_gen_helper(100, 2)
test_ron_gauss_instance = ron_gauss.RONGauss(algorithm="unsupervised")
dp_data = test_ron_gauss_instance.generate_dpdata(test_data, 1)
assert normality_test_helper(dp_data)
def test_defaults():
test_ron_gauss_instance = ron_gauss.RONGauss()
assert (test_ron_gauss_instance.algorithm == "supervised")
assert (test_ron_gauss_instance.epsilonCov == 1.0)
assert (test_ron_gauss_instance.epsilonMean == 1.0)