def test_compute_affinity_matrix_args(almost_equal_decimals=5):
""" test the compute_affinity_matrix parameter arguments """
input_types = ['data', 'adjacency', 'affinity']
params = [{'radius': 4}, {'radius': 5}]
adjacency_method = 'auto'
for affinity_method in affinity_methods():
X = np.random.uniform(size=(10, 2))
D = compute_adjacency_matrix(X, adjacency_method, **params[1])
A = compute_affinity_matrix(D, affinity_method, **params[1])
for init_params in params:
for kwarg_params in params:
true_params = init_params.copy()
true_params.update(kwarg_params)
affinity_true = compute_affinity_matrix(
D, adjacency_method, **true_params)
for input in input_types:
G = Geometry(adjacency_method=adjacency_method,
adjacency_kwds=params[1],
affinity_method=affinity_method,
affinity_kwds=init_params)
if input in ['data', 'adjacency']:
if input in ['data']:
G.set_data_matrix(X)
else:
G.set_adjacency_matrix(D)
affinity_queried = G.compute_affinity_matrix(
**kwarg_params)
assert_array_almost_equal(affinity_true.todense(),
affinity_queried.todense(),
almost_equal_decimals)
else:
G.set_affinity_matrix(A)
msg = distance_error_msg
assert_raise_message(ValueError, msg,
G.compute_affinity_matrix)
def test_compute_affinity_matrix_args(almost_equal_decimals=5):
""" test the compute_affinity_matrix parameter arguments """
input_types = ['data', 'adjacency', 'affinity']
params = [{'radius':4}, {'radius':5}]
adjacency_method = 'auto'
for affinity_method in affinity_methods():
X = np.random.uniform(size=(10, 2))
D = compute_adjacency_matrix(X, adjacency_method, **params[1])
A = compute_affinity_matrix(D, affinity_method, **params[1])
for init_params in params:
for kwarg_params in params:
true_params = init_params.copy()
true_params.update(kwarg_params)
affinity_true = compute_affinity_matrix(D, adjacency_method,
**true_params)
for input in input_types:
G = Geometry(adjacency_method = adjacency_method,
adjacency_kwds = params[1],
affinity_method = affinity_method,
affinity_kwds = init_params)
if input in ['data', 'adjacency']:
if input in ['data']:
G.set_data_matrix(X)
else:
G.set_adjacency_matrix(D)
affinity_queried = G.compute_affinity_matrix(**kwarg_params)
assert_array_almost_equal(affinity_true.todense(), affinity_queried.todense(), almost_equal_decimals)
else:
G.set_affinity_matrix(A)
msg = distance_error_msg
assert_raise_message(ValueError, msg, G.compute_affinity_matrix)
def test_affinity_vs_matlab():
"""Test that the affinity calculation matches the matlab result"""
matlab = io.loadmat(TEST_DATA)
D = np.sqrt(matlab['S']) # matlab outputs squared distances
A_matlab = matlab['A']
radius = matlab['rad'][0]
# check dense affinity computation
A_dense = compute_affinity_matrix(D, radius=radius)
assert_allclose(A_dense, A_matlab)
# check sparse affinity computation
A_sparse = compute_affinity_matrix(csr_matrix(D), radius=radius)
assert_allclose(A_sparse.toarray(), A_matlab)
def test_compute_laplacian_matrix_args(almost_equal_decimals=5):
input_types = ['data', 'adjacency', 'affinity']
params = [{}, {'radius': 4}, {'radius': 5}]
lapl_params = [{}, {'scaling_epps': 4}, {'scaling_epps': 10}]
adjacency_method = 'auto'
affinity_method = 'auto'
for laplacian_method in laplacian_methods():
X = np.random.uniform(size=(10, 2))
D = compute_adjacency_matrix(X, adjacency_method, **params[1])
A = compute_affinity_matrix(D, affinity_method, **params[1])
for init_params in lapl_params:
for kwarg_params in lapl_params:
true_params = init_params.copy()
true_params.update(kwarg_params)
laplacian_true = compute_laplacian_matrix(
A, laplacian_method, **true_params)
for input in input_types:
G = Geometry(adjacency_method=adjacency_method,
adjacency_kwds=params[1],
affinity_method=affinity_method,
affinity_kwds=params[1],
laplacian_method=laplacian_method,
laplacian_kwds=lapl_params[0])
if input in ['data']:
G.set_data_matrix(X)
if input in ['adjacency']:
G.set_adjacency_matrix(D)
else:
G.set_affinity_matrix(A)
laplacian_queried = G.compute_laplacian_matrix(**kwarg_params)
assert_array_almost_equal(laplacian_true.todense(),
laplacian_queried.todense(),
almost_equal_decimals)
def test_compute_laplacian_matrix_args(almost_equal_decimals=5):
input_types = ['data', 'adjacency', 'affinity']
params = [{}, {'radius':4}, {'radius':5}]
lapl_params = [{}, {'scaling_epps':4}, {'scaling_epps':10}]
adjacency_method = 'auto'
affinity_method = 'auto'
for laplacian_method in laplacian_methods():
X = np.random.uniform(size=(10, 2))
D = compute_adjacency_matrix(X, adjacency_method, **params[1])
A = compute_affinity_matrix(D, affinity_method, **params[1])
for init_params in lapl_params:
for kwarg_params in lapl_params:
true_params = init_params.copy()
true_params.update(kwarg_params)
laplacian_true = compute_laplacian_matrix(A, laplacian_method, **true_params)
for input in input_types:
G = Geometry(adjacency_method = adjacency_method,
adjacency_kwds = params[1],
affinity_method = affinity_method,
affinity_kwds = params[1],
laplacian_method = laplacian_method,
laplacian_kwds = lapl_params[0])
if input in ['data']:
G.set_data_matrix(X)
if input in ['adjacency']:
G.set_adjacency_matrix(D)
else:
G.set_affinity_matrix(A)
laplacian_queried = G.compute_laplacian_matrix(**kwarg_params)
assert_array_almost_equal(laplacian_true.todense(), laplacian_queried.todense(), almost_equal_decimals)
def test_compute_adjacency_matrix_args(almost_equal_decimals=5):
""" test the compute_adjacency_matrix parameter arguments """
input_types = ['data', 'adjacency', 'affinity']
params = [{'radius':1}, {'radius':2}]
for adjacency_method in adjacency_methods():
if adjacency_method == 'pyflann':
try:
import pyflann as pyf
except ImportError:
raise SkipTest("pyflann not installed.")
X = np.random.uniform(size=(10, 2))
D = compute_adjacency_matrix(X, adjacency_method, **params[1])
A = compute_affinity_matrix(D, radius=1)
for init_params in params:
for kwarg_params in params:
true_params = init_params.copy()
true_params.update(kwarg_params)
adjacency_true = compute_adjacency_matrix(X, adjacency_method, **true_params)
G = Geometry(adjacency_method=adjacency_method, adjacency_kwds=init_params)
for input in input_types:
G = Geometry(adjacency_kwds = init_params)
if input in ['data']:
G.set_data_matrix(X)
adjacency_queried = G.compute_adjacency_matrix(**kwarg_params)
assert_allclose(adjacency_true.toarray(),
adjacency_queried.toarray(), rtol=1E-5)
else:
if input in ['adjacency']:
G.set_adjacency_matrix(D)
else:
G.set_affinity_matrix(A)
msg = distance_error_msg
assert_raise_message(ValueError, msg, G.compute_adjacency_matrix)
def check_affinity(adjacency_radius, affinity_radius, symmetrize):
adj = compute_adjacency_matrix(X, radius=adjacency_radius)
aff = compute_affinity_matrix(adj, radius=affinity_radius,
symmetrize=True)
A = np.exp(-(D / affinity_radius) ** 2)
A[D > adjacency_radius] = 0
assert_allclose(aff.toarray(), A)
def check_symmetric(method, adjacency_radius, affinity_radius):
adj = compute_adjacency_matrix(X, radius=adjacency_radius)
aff = compute_affinity_matrix(adj, radius=affinity_radius)
lap, lapsym, w = compute_laplacian_matrix(aff, method=method,
full_output=True)
sym = w[:, np.newaxis] * (lap.toarray() + np.eye(*lap.shape))
assert_allclose(lapsym.toarray(), sym)
def test_affinity_sparse_vs_dense():
"""
Test that A_sparse is the same as A_dense for a small A matrix
"""
rad = 2.
n_samples = 6
X = np.arange(n_samples)
X = X[ :,np.newaxis]
X = np.concatenate((X,np.zeros((n_samples,1),dtype=float)),axis=1)
X = np.asarray( X, order="C" )
test_dist_matrix = compute_adjacency_matrix( X, method = 'auto', radius = rad )
A_dense = compute_affinity_matrix(test_dist_matrix.toarray(), method = 'auto',
radius = rad, symmetrize = False )
A_sparse = compute_affinity_matrix(csr_matrix(test_dist_matrix),
method = 'auto', radius = rad, symmetrize = False)
A_spdense = A_sparse.toarray()
A_spdense[ A_spdense == 0 ] = 1.
assert_allclose(A_dense, A_spdense)
def check_symmetric(method, adjacency_radius, affinity_radius):
adj = compute_adjacency_matrix(X, radius=adjacency_radius)
aff = compute_affinity_matrix(adj, radius=affinity_radius)
lap, lapsym, w = compute_laplacian_matrix(aff,
method=method,
full_output=True)
sym = w[:, np.newaxis] * (lap.toarray() + np.eye(*lap.shape))
assert_allclose(lapsym.toarray(), sym)
def check_laplacian(input_type, laplacian_method):
kwargs = {'scaling_epps': radius}
if laplacian_method == 'renormalized':
kwargs['renormalization_exponent'] = 1.5
adjacency = input_type(np.sqrt(matlab['S']))
affinity = compute_affinity_matrix(adjacency, radius=radius)
laplacian = compute_laplacian_matrix(affinity,
method=laplacian_method,
**kwargs)
if input_type is csr_matrix:
laplacian = laplacian.toarray()
assert_allclose(laplacian, laplacians[laplacian_method])
def check_laplacian(input_type, laplacian_method):
kwargs = {'scaling_epps': radius}
if laplacian_method == 'renormalized':
kwargs['renormalization_exponent'] = 1.5
adjacency = input_type(np.sqrt(matlab['S']))
affinity = compute_affinity_matrix(adjacency, radius=radius)
laplacian = compute_laplacian_matrix(affinity,
method=laplacian_method,
**kwargs)
if input_type is csr_matrix:
laplacian = laplacian.toarray()
assert_allclose(laplacian, laplacians[laplacian_method])
def test_custom_affinity():
class CustomAffinity(Affinity):
name = "custom"
def affinity_matrix(self, adjacency_matrix):
return np.exp(-abs(adjacency_matrix.toarray()))
rand = np.random.RandomState(42)
X = rand.rand(10, 2)
D = compute_adjacency_matrix(X, radius=10)
A = compute_affinity_matrix(D, method='custom', radius=1)
assert_allclose(A, np.exp(-abs(D.toarray())))
Affinity._remove_from_registry("custom")
def test_laplacian_smoketest():
rand = np.random.RandomState(42)
X = rand.rand(20, 2)
adj = compute_adjacency_matrix(X, radius=0.5)
aff = compute_affinity_matrix(adj, radius=0.1)
def check_laplacian(method):
lap = compute_laplacian_matrix(aff, method=method)
assert isspmatrix(lap)
assert_equal(lap.shape, (X.shape[0], X.shape[0]))
for method in Laplacian.asymmetric_methods():
yield check_laplacian, method
def test_laplacian_smoketest():
rand = np.random.RandomState(42)
X = rand.rand(20, 2)
adj = compute_adjacency_matrix(X, radius=0.5)
aff = compute_affinity_matrix(adj, radius=0.1)
def check_laplacian(method):
lap = compute_laplacian_matrix(aff, method=method)
assert isspmatrix(lap)
assert_equal(lap.shape, (X.shape[0], X.shape[0]))
for method in Laplacian.asymmetric_methods():
yield check_laplacian, method
def test_compute_adjacency_matrix_args(almost_equal_decimals=5):
""" test the compute_adjacency_matrix parameter arguments """
input_types = ['data', 'adjacency', 'affinity']
params = [{'radius': 1}, {'radius': 2}]
for adjacency_method in adjacency_methods():
if adjacency_method == 'pyflann':
try:
import pyflann as pyf
except ImportError:
raise SkipTest("pyflann not installed.")
X = np.random.uniform(size=(10, 2))
D = compute_adjacency_matrix(X, adjacency_method, **params[1])
A = compute_affinity_matrix(D, radius=1)
for init_params in params:
for kwarg_params in params:
true_params = init_params.copy()
true_params.update(kwarg_params)
adjacency_true = compute_adjacency_matrix(
X, adjacency_method, **true_params)
G = Geometry(adjacency_method=adjacency_method,
adjacency_kwds=init_params)
for input in input_types:
G = Geometry(adjacency_kwds=init_params)
if input in ['data']:
G.set_data_matrix(X)
adjacency_queried = G.compute_adjacency_matrix(
**kwarg_params)
assert_allclose(adjacency_true.toarray(),
adjacency_queried.toarray(),
rtol=1E-5)
else:
if input in ['adjacency']:
G.set_adjacency_matrix(D)
else:
G.set_affinity_matrix(A)
msg = distance_error_msg
assert_raise_message(ValueError, msg,
G.compute_adjacency_matrix)
