def test_correct_num_features(data):
"""Test the exact number of features are selected."""
for num_features in [2, 3, 4]:
dgufs = DGUFS(num_features=num_features)
dgufs.fit(data)
assert len(dgufs.support) == num_features
def test_iris(iris):
"""Select the two most important features from the Iris data set."""
dgufs = DGUFS(num_features=2)
dgufs.fit(iris)
assert len(dgufs.support) == 2
def test_V(data):
"""Test (n clusters x nrows) = shape(V)."""
nrows, _ = np.shape(data)
dgufs = DGUFS()
dgufs.fit(data)
assert np.shape(dgufs.cluster_labels) == (dgufs.num_clusters, nrows)
def test_L(data):
""" Test properties of L matrix."""
dgufs = DGUFS()
dgufs.fit(data)
# Test L is positive semi-definite.
assert np.all(np.linalg.eigvals(dgufs.L) >= 0)
def test_Y_norm(data):
# Test norm(Y)_(2, 0) = m
dgufs = DGUFS()
dgufs.fit(data)
l20_norm = np.linalg.norm(np.linalg.norm(dgufs.Y, ord=2, axis=1), ord=0)
assert l20_norm == dgufs.num_features
def test_M(data):
"""Test M matrix is symmetric and binary."""
nrows, _ = np.shape(data)
dgufs = DGUFS()
dgufs.fit(data)
assert np.shape(dgufs.M) == (nrows, nrows)
assert len(np.unique(dgufs.M)) == 2
def test_error_num_features(data):
"""Test an error is raised if specifying too many features."""
with pytest.raises(ValueError):
dgufs = DGUFS(num_features=10)
dgufs.fit(data)