def test_quotient_bad_partition():
with pytest.raises(ValueError):
quotient_similarity(m, partition=[[0, 2], [3, 7]], check=True)
def test_nx():
mq1 = quotient_similarity(m, partition, agg='sum',
diag_value=0).todense()
Gq = quotient_graph(G=G, partition=partition, edge_data=edge_sum)
mq2 = to_numpy_array(Gq)
assert allclose(mq1, mq2)
def test_quotient_custom():
assert allclose(
quotient_similarity(m, partition, agg=custom_agg, n_cpu=-1).todense(),
m_merged)
def test_quotient_parallel():
assert allclose(
quotient_similarity(m, partition, agg='sum', n_cpu=-1).todense(),
m_merged)
def test_quotient_wrong_agg_else():
with pytest.raises(ValueError):
quotient_similarity(m, partition, agg=array([]), n_cpu=1)
def test_quotient_wrong_agg_str():
with pytest.raises(ValueError):
quotient_similarity(m, partition, agg='ABCD', n_cpu=1)
m[m < sim_thresh] = 0
m = csr_matrix(m)
# Generate IDs for partition
ids = list(range(int(sqrt(r))))
id_map = array(choices(ids, k=r))
partition = [where(id_map == i)[0].tolist() for i in ids]
# Networkx implementation
start_time = timeit.default_timer()
nx_G = nx.from_scipy_sparse_matrix(m)
nx_Gq = nx.quotient_graph(G=nx_G, partition=partition)
nx_q = nx.to_scipy_sparse_matrix(nx_Gq)
nx_q.eliminate_zeros()
nx_time.append(timeit.default_timer() - start_time)
# PSS implementation
start_time = timeit.default_timer()
pss_q = quotient_similarity(m, partition, diag_value=0)
pss_time.append(timeit.default_timer() - start_time)
# Test and print
if not allclose(nx_q.todense(), pss_q.todense()):
raise ValueError('Quotient similarity matrix values diverged')
print('END {} rows'.format(r))
plt.plot(n_rows,
nx_time,
color='black',
linestyle='dashed',
linewidth=2,
label='Networkx')
plt.plot(n_rows, pss_time, color='green', linewidth=2, label='pysimscale')
plt.title('Quotient Similarity Graph')
plt.legend()
