def test_kernels():
rs = np.random.RandomState(0)
x = rs.randn(10, 15)
a = np.corrcoef(x)
# with pytest.warns(UserWarning):
a2 = compute_affinity(a, sparsity=None)
assert np.count_nonzero(a2 < 0) == 0
a2 = compute_affinity(a, sparsity=.7)
assert np.count_nonzero(a2 < 0) == 0
assert np.all(np.count_nonzero(a2, axis=1) == 3)
a2 = compute_affinity(a, kernel='cosine', sparsity=.7)
assert np.count_nonzero(a2 < 0) == 0
def test_embedding_gradient():
rs = np.random.RandomState(0)
x = rs.randn(100, 50)
x2 = rs.randn(100, 50)
a = compute_affinity(x, kernel='gaussian', sparsity=0.7)
a_sparse = coo_matrix(a)
dop = {
'dm': emb.DiffusionMaps,
'le': emb.LaplacianEigenmaps,
'pca': emb.PCAMaps
}
for app_name, app in dop.items():
# test dense
m = app(random_state=0)
m.fit(a)
assert m.lambdas_.shape == (10, )
assert m.maps_.shape == (100, 10)
if app_name == 'le':
assert np.allclose(m.lambdas_, np.sort(m.lambdas_))
else:
assert np.allclose(m.lambdas_, np.sort(m.lambdas_)[::-1])
# test sparse
m2 = app(random_state=0)
if app_name == 'pca':
with pytest.raises(Exception):
m2.fit(a_sparse)
else:
m2.fit(a_sparse)
assert np.allclose(m.lambdas_, m2.lambdas_)
assert np.allclose(m.maps_, m2.maps_)
# test with gradientmaps
gm = GradientMaps(approach=app_name, kernel='gaussian', random_state=0)
gm.fit(x, sparsity=0.7)
assert np.allclose(gm.lambdas_, m.lambdas_)
assert np.allclose(gm.gradients_, m.maps_)
assert gm.aligned_ is None
# test alignment
for align in [None, 'procrustes', 'joint']:
gm_dm = GradientMaps(approach='dm',
kernel='gaussian',
alignment=align,
random_state=0)
gm_dm.fit([x, x2], sparsity=0.7)
assert len(gm_dm.gradients_) == 2
assert len(gm_dm.lambdas_) == 2
if align is None:
assert gm_dm.aligned_ is None
elif align == 'procrutes':
for i in range(2):
assert not np.all(gm_dm.aligned_[i] == gm_dm.gradients_[i])
elif align == 'joint':
for i in range(2):
assert np.all(gm_dm.aligned_[i] == gm_dm.gradients_[i])
for i in TD_index:
conn_mat_TD.append(np.load(file_list[i]))
conn_mat_ASD = np.array(conn_mat_ASD)
conn_mat_TD = np.array(conn_mat_TD)
conn_mat = np.concatenate((conn_mat_ASD, conn_mat_TD))
print('conn_mat_ASD : ', conn_mat_ASD.shape, ' ', 'conn_mat_TD : ',
conn_mat_TD.shape, ' ', 'conn_mat : ', conn_mat.shape)
# # Make affine matrix
affine_conn_mat = []
for i, x in enumerate(conn_mat):
print(i, ' ', end='', flush=True)
z_conn_mat = np.nan_to_num(np.arctanh(np.nan_to_num(x, nan=0.0)), nan=0.0)
noaff_conn_mat = gradient.compute_affinity(z_conn_mat, sparsity=0.7)
affine_conn_mat.append(noaff_conn_mat)
affine_conn_mat = np.array(affine_conn_mat)
np.save(savepath)
grpmean_conn_mat = np.nan_to_num(np.arctanh(
np.nan_to_num(np.mean(conn_mat, axis=0), nan=0.0)),
nan=0.0)
noaff_grpmean_conn_mat = gradient.compute_affinity(
grpmean_conn_mat, sparsity=0.7) # sparsity로 thresholding ratio 조절
noaff_grpmean_conn_mat = np.nan_to_num(noaff_grpmean_conn_mat, nan=0.0)
grpmean_conn_mat = noaff_grpmean_conn_mat
np.save(savepath)