def test_cosine_convertable_dtype():
assert allclose(
truncated_sparse_similarity(a1.astype('object'),
metric='cosine',
thresh=0,
diag_value=None,
n_jobs=1).todense(), expected_cosine_sim)
def test_hamming_no_thresh_dense():
assert allclose(
truncated_sparse_similarity(a2,
metric='hamming',
thresh=0,
diag_value=0,
n_jobs=1).todense(), expected_hamming_sim)
def test_cosine_no_thresh():
assert allclose(
truncated_sparse_similarity(a1,
metric='cosine',
thresh=0,
diag_value=None,
n_jobs=1).todense(), expected_cosine_sim)
def test_sim_sparsity():
assert issparse(
truncated_sparse_similarity(a1,
metric='cosine',
thresh=0.9,
diag_value=0,
n_jobs=1))
def test_cosine_thresh():
sim_thresh = 0.9
expected = expected_cosine_sim.copy()
expected[expected < sim_thresh] = 0
assert allclose(
truncated_sparse_similarity(a1,
metric='cosine',
thresh=sim_thresh,
diag_value=None,
n_jobs=1).todense(), expected)
def test_cosine_parallel():
if HAS_JOBLIB:
assert allclose(
truncated_sparse_similarity(a1,
metric='cosine',
thresh=0,
diag_value=None,
n_jobs=-1).todense(),
expected_cosine_sim)
else:
print('Could not find a Joblib instalation, skipping test')
def test_hamming_thresh_dense():
sim_thresh = 0.75
expected = expected_hamming_sim.copy()
expected[expected < sim_thresh] = 0
if HAS_JOBLIB:
n_cpu = -1
else:
n_cpu = 1
assert allclose(
truncated_sparse_similarity(a2,
metric='hamming',
thresh=sim_thresh,
diag_value=0,
n_jobs=n_cpu).todense(), expected)
sk_time = []
pss_row_time = []
pss_block_time = []
for r in n_rows:
m = normal(0, 1, (r, n_features))
start_time = timeit.default_timer()
sk = cosine_similarity(m, dense_output=False)
sk[sk < sim_thresh] = 0
sk = csr_matrix(sk)
sk.setdiag(0)
sk.eliminate_zeros()
sk_time.append(timeit.default_timer() - start_time)
start_time = timeit.default_timer()
pss_row = truncated_sparse_similarity(m,
metric='cosine',
thresh=sim_thresh,
block_size=1)
pss_row_time.append(timeit.default_timer() - start_time)
start_time = timeit.default_timer()
pss_block = truncated_sparse_similarity(m,
metric='cosine',
thresh=sim_thresh,
block_size=1000)
pss_block_time.append(timeit.default_timer() - start_time)
print('END {} rows'.format(r))
plt.plot(n_rows,
sk_time,
color='black',
linestyle='dashed',
linewidth=2,
def test_cosine_bad_dtype():
with pytest.raises(TypeError):
truncated_sparse_similarity(array([lambda x: x, lambda y: y]))
def test_sim_wrong_metric():
with pytest.raises(ValueError):
truncated_sparse_similarity(a1, metric=None, n_jobs=1)