def main(self,
algo,
param,
privacy,
e_iter=100000,
d_iter=500000,
test_range=0.1,
n_checks=3):
# list of tasks to test, each tuple contains (function, extra_args, sensitivity)
tasks = [(generic_method_pydp, {
'algorithm': algo,
'param_for_algorithm': param
}, ALL_DIFFER)]
# claimed privacy level to check
claimed_privacy = privacy
for i, (algorithm, kwargs, sensitivity) in enumerate(tasks):
start_time = time.time()
results = {}
flag_file = time.ctime().replace(' ', '_')
for privacy_budget in claimed_privacy:
# # privacy levels to test, here we test the claimed privacy plus .01 above and below
test_privacy = arr_n_check(privacy_budget, test_range,
n_checks)
# set the second argument of the function (assumed to be `epsilon`) to the claimed privacy level
kwargs[algorithm.__code__.co_varnames[1]] = privacy_budget
results[privacy_budget] = self.detect_counterexample(
algorithm,
test_privacy,
kwargs,
sensitivity=sensitivity,
event_iterations=e_iter,
detect_iterations=d_iter)
# dump the results to file
json_file = Path.cwd() / f'{algorithm.__name__}_{flag_file}.json'
with json_file.open('w') as f:
json.dump(encode(results, unpicklable=False), f)
# plot and save to file
plot_file = Path.cwd() / f'{algorithm.__name__}_{flag_file}.pdf'
self.plot_result(results, r'Test $\epsilon$', 'P Value',
algorithm.__name__.replace('_', ' ').title(),
plot_file)
total_time, total_detections = time.time() - start_time, len(claimed_privacy) * \
len(test_privacy)
logger.info(
f'[{i + 1} / {len(tasks)}]: {algorithm.__name__} | Time elapsed: {total_time:5.3f}s | '
f'Average time per detection: {total_time / total_detections:5.3f}s'
)
return 0
def test_negative_privacy():
assert arr_n_check(-0.9, 0.5, 3) == (0.4, 0.9, 1.4)
def test_very_large_test_range():
assert arr_n_check(0.9, 1, 5) == (0.9, )
def test_inappropriats_n_checks1():
assert arr_n_check(0.9, 0.5, 4) == (0.4, 0.9, 1.4)