def test_permutation(requests, the_truth):
project_data, (r_a, r_b) = create_project_upload_data(
requests, (the_truth['clks_a'], the_truth['clks_b']),
result_type='permutations')
run = post_run(requests, project_data, threshold=the_truth['threshold'])
mask_result = get_run_result(requests, project_data, run, timeout=240)
perm_a_result = get_run_result(requests,
project_data,
run,
result_token=r_a['receipt_token'],
wait=False)
perm_b_result = get_run_result(requests,
project_data,
run,
result_token=r_b['receipt_token'],
wait=False)
# compare permutations and mask against mapping of the truth
permutation_a = inverse_of_permutation(perm_a_result['permutation'])
permutation_b = inverse_of_permutation(perm_b_result['permutation'])
groups = the_truth['groups']
# Use a mapping output to simplify the checking.
mapping = dict(anonlink.solving.pairs_from_groups(groups))
# NB: Anonlink is more strict on enforcing the k parameter, so there
# is a small chance the below won't hold. This should only be the
# case for more noisy problems.
for a, b, m in zip(permutation_a, permutation_b, mask_result['mask']):
if m == 1:
assert a in mapping, f"Unexpected link was included - run {run}"
assert mapping[
a] == b, f"Expected link from {a} was incorrect - run {run}"
else:
assert a not in mapping, f"Expected link was masked out - run {run}"
def test_groups_correctness(requests):
# We assume that anonlink computes the right results.
with open(DATA_PATH, 'rb') as f:
# Here's some filters I prepared earlier.
filters = pickle.load(f)
candidate_pairs = anonlink.candidate_generation.find_candidate_pairs(
filters, anonlink.similarities.dice_coefficient_accelerated, THRESHOLD)
true_groups = anonlink.solving.greedy_solve(candidate_pairs)
filter_size = len(filters[0][0])
assert all(
len(filter_) == filter_size for dataset in filters
for filter_ in dataset)
packed_filters = [
b''.join(binary_pack_for_upload(f, filter_size)) for f in filters
]
project_data, _ = create_project_upload_data(requests,
packed_filters,
result_type='groups',
binary=True,
hash_size=DATA_HASH_SIZE)
try:
run = post_run(requests, project_data, threshold=THRESHOLD)
result_groups = get_run_result(requests, project_data, run)['groups']
finally:
delete_project(requests, project_data)
# Compare ES result with anonlink.
result_group_set = {frozenset(map(tuple, g)) for g in result_groups}
true_group_set = set(map(frozenset, true_groups))
assert result_group_set == true_group_set
def test_project_json_data_upload_with_mismatched_encoded_size(
requests, result_type_number_parties):
result_type, number_parties = result_type_number_parties
data = [generate_json_serialized_clks(500, 64 if i == 0 else 256)
for i in range(number_parties)]
new_project_data, _ = create_project_upload_data(
requests, data, result_type=result_type)
with pytest.raises(AssertionError):
run_id = post_run(requests, new_project_data, 0.9)
get_run_result(requests, new_project_data, run_id, wait=True)
def test_groups(requests, the_truth):
project_data, _ = create_project_upload_data(
requests, (the_truth['clks_a'], the_truth['clks_b']),
result_type='groups')
run = post_run(requests, project_data, threshold=the_truth['threshold'])
result = get_run_result(requests, project_data, run, timeout=240)
# compare mapping with the truth
result_groups = result['groups']
true_groups = the_truth['groups']
result_groups = frozenset(
frozenset(map(tuple, group)) for group in result_groups)
true_groups = frozenset(map(frozenset, true_groups))
assert result_groups == true_groups
def test_similarity_scores(requests, the_truth):
project_data, _ = create_project_upload_data(
requests,
(the_truth['clks_a'], the_truth['clks_b']),
result_type='similarity_scores')
run = post_run(requests, project_data, threshold=the_truth['threshold'])
result = get_run_result(requests, project_data, run, timeout=60)
true_scores = the_truth['similarity_scores']
result_scores = {tuple(index for _, index in sorted([a, b])): score
for a, b, score in result['similarity_scores']}
# Anonlink is more strict on enforcing the k parameter. Hence the
# subset.
assert true_scores.keys() <= result_scores.keys()
for pair in true_scores:
assert true_scores[pair] == result_scores[pair]
delete_project(requests, project_data)