def test_parallel_get_pairwise_comparisons(neighbors_object_list, data, metric):
neighbors_list = []
n_neighbors_list = []
for neighbors_object in neighbors_object_list:
neighbors_list.append(neighbors_object.neighbors)
n_neighbors_list.append(neighbors_object.n_neighbors)
n_neighbors = data.draw(st.integers(min_value=2, max_value=min(n_neighbors_list)))
# set parameters so we don't hang on small examples
num_comparisons = len(list(embed.knn_iterator(neighbors_list)))
if num_comparisons < 2000:
n_proc = 1
chunksize = None
else:
n_proc = 2
chunksize = 1000
comparisons = embed.get_pairwise_comparisons(neighbors_list, n_neighbors=n_neighbors, how=metric)
p_comparisons = embed.parallel_get_pairwise_comparisons(neighbors_list, n_neighbors=n_neighbors, how=metric, n_proc=n_proc, chunksize=chunksize)
for comparison in comparisons:
i1, i2 = comparison[0], comparison[1]
for i in range(len(p_comparisons)):
if i1 == p_comparisons[i][0] and i2 == p_comparisons[i][1]:
assert(comparison_equality(comparison[2], p_comparisons[i][2]))
def test_parallel_get_pairwise_comparisons_varying_k(neighbors_object_list, data, metric):
neighbors_list = []
n_neighbors_list = []
for neighbors_object in neighbors_object_list:
neighbors_list.append(neighbors_object.neighbors)
n_neighbors_list.append(neighbors_object.n_neighbors)
n_neighbors = data.draw(st.integers(min_value=2, max_value=min(n_neighbors_list)))
max_k = data.draw(st.integers(min_value=2, max_value=n_neighbors))
# set parameters so we don't hang on small examples
num_comparisons = len(list(embed.knn_iterator(neighbors_list, max_k)))
if num_comparisons < 2000:
n_proc = 1
chunksize = None
else:
n_proc = 2
chunksize = 1000
comparisons = embed.parallel_get_pairwise_comparisons_varying_k(neighbors_list, max_k, how=metric, n_proc=n_proc, chunksize=chunksize)
for comparison in comparisons:
i1, i2, k = comparison[0], comparison[1], comparison[2]
test_comparison = embed.compare_neighbors(neighbors_list[i1][0], neighbors_list[i2][0], n_neighbors=k, how=metric)
assert(comparison_equality(comparison[3], test_comparison))
def test_knn_iterate(neighbors_object_list, data):
n_neighbors_list = []
neighbors_list = []
for neighbors_object in neighbors_object_list:
n_neighbors_list.append(neighbors_object.n_neighbors)
neighbors_list.append(neighbors_object.neighbors)
n_neighbors = data.draw(st.integers(min_value=2, max_value=min(n_neighbors_list)))
knn_iter = embed.knn_iterator(neighbors_list, n_neighbors=n_neighbors)
knn_iter_list = list(knn_iter)
# compute the expected length of knn_iter_list
pairs = [(a, b) for a, b in comb(range(len(neighbors_list)), 2) if a < b]
knn_length = 0
for a, b in pairs:
knn_dict_a, _ = neighbors_list[a]
knn_dict_b, _ = neighbors_list[b]
knn_length += len(set(knn_dict_a.keys()).intersection(knn_dict_b.keys()))
assert(knn_length == len(knn_iter_list))
# check elements satisfy basic expected properties
for item in knn_iter_list:
(i1, i2), word, n1, n2 = item
assert(i1 < i2)
note(neighbors_list[i1][0][word])
note(n1)
assert(neighbors_list[i1][0][word][1:n_neighbors] == n1)
assert(neighbors_list[i2][0][word][1:n_neighbors] == n2)
def test_aggregate_comparisons_stats(neighbors_object_list, data, metric):
basefilename = "test_comparisons"
data_path = test_dir
neighbors_list = []
n_neighbors_list = []
metadata_list = []
for neighbors_object in neighbors_object_list:
neighbors_list.append(neighbors_object.neighbors)
n_neighbors_list.append(neighbors_object.n_neighbors)
metadata_list.append(neighbors_object.metadata)
n_neighbors = data.draw(st.integers(min_value=2, max_value=min(n_neighbors_list)))
# test get_pairwise_comparisons output
comparisons = embed.get_pairwise_comparisons(neighbors_list, n_neighbors=n_neighbors, how=metric)
results, results_mean, results_variance = embed.aggregate_comparison_stats(comparisons)
for i, j, (labels, similarity) in comparisons:
# check symmetry
assert(results_mean[i, j] == results_mean[j, i])
assert(results_variance[i, j] == results_variance[j, i])
describe = scipy.stats.describe(similarity)
assert(describe == results[i, j])
assert(results_mean[i, j] == scipy.mean(similarity))
assert(describe.variance == results_variance[i, j])
# test on parallel_get_pairwise_comparisons_varying_k output
max_k = data.draw(st.integers(min_value=2, max_value=n_neighbors))
# set parameters so we don't hang on small examples
num_comparisons = len(list(embed.knn_iterator(neighbors_list, max_k)))
if num_comparisons < 2000:
n_proc = 1
chunksize = None
else:
n_proc = 2
chunksize = 1000
comparisons = embed.parallel_get_pairwise_comparisons_varying_k(neighbors_list, max_k,
n_proc=n_proc, chunksize=chunksize, how=metric)
results, results_mean, results_variance = embed.aggregate_comparison_stats(comparisons, max_k=max_k)
for i, j, k, (labels, similarity) in comparisons:
# check symmetry
assert(k <= max_k)
assert(results_mean[i, j, k] == results_mean[j, i, k])
assert(results_variance[i, j, k] == results_variance[j, i, k])
describe = scipy.stats.describe(similarity)
assert(describe == results[i, j, k])
assert(results_mean[i, j, k] == scipy.mean(similarity))
assert(describe.variance == results_variance[i, j, k])
def test_save_and_read_comparisons(neighbors_object_list, data, metric):
basefilename = "test_comparisons"
data_path = test_dir
neighbors_list = []
n_neighbors_list = []
metadata_list = []
for neighbors_object in neighbors_object_list:
neighbors_list.append(neighbors_object.neighbors)
n_neighbors_list.append(neighbors_object.n_neighbors)
metadata_list.append(neighbors_object.metadata)
n_neighbors = data.draw(st.integers(min_value=2, max_value=min(n_neighbors_list)))
# test out return format of get_pairwise_comparisons
comparisons = embed.get_pairwise_comparisons(neighbors_list, n_neighbors=n_neighbors, how=metric)
embed.save_comparisons(basefilename, comparisons=comparisons, metadata=metadata_list, data_path=data_path)
read_comparisons, read_metadata = embed.read_comparisons(basefilename, data_path=test_dir)
comparisons_filename = os.path.join(data_path, basefilename + ".comparisons")
metadata_filename = os.path.join(data_path, basefilename + ".comparisons_metadata")
os.remove(comparisons_filename)
os.remove(metadata_filename)
assert(comparisons == read_comparisons)
assert(metadata_list == read_metadata)
# test save/read from parallel_get_pairwise_comparisons_varying_k
max_k = data.draw(st.integers(min_value=2, max_value=n_neighbors))
# set parameters so we don't hang on small examples
num_comparisons = len(list(embed.knn_iterator(neighbors_list, max_k)))
if num_comparisons < 2000:
n_proc = 1
chunksize = None
else:
n_proc = 2
chunksize = 1000
comparisons = embed.parallel_get_pairwise_comparisons_varying_k(neighbors_list, max_k,
n_proc=n_proc, chunksize=chunksize, how=metric)
embed.save_comparisons(basefilename, max_k=max_k, comparisons=comparisons, metadata=metadata_list, data_path=data_path)
read_comparisons, read_metadata = embed.read_comparisons(basefilename, data_path=test_dir)
comparisons_filename = os.path.join(data_path, basefilename + ".comparisons")
metadata_filename = os.path.join(data_path, basefilename + ".comparisons_metadata")
os.remove(comparisons_filename)
os.remove(metadata_filename)
assert(comparisons == read_comparisons)
assert(metadata_list == read_metadata)