def test_tolerance(self):
"""
Tests that tolerance truncates the eigendecomposition
"""
tol = 1e-3
v, _ = eig_solver(self.X, tol=tol)
self.assertGreaterEqual(max(v), tol)
def test_no_ncomponents(self):
"""
Tests that if n_components is None and tol is None,
returns full eigendecomposition
"""
v, U = eig_solver(self.X, tol=None)
self.assertTrue(
len(v) == self.X.shape[0] and U.shape[0] == self.X.shape[0])
def test_add_null(self):
"""
Tests that null eigenvectors / values are returned when n_components is
greater than the number of significant eigenpairs
"""
v, _ = np.linalg.eig(self.X)
tol = 1e-3
n_eig = len(v[v >= tol])
n_components = self.X.shape[0]
v, U = eig_solver(self.X,
n_components=n_components,
tol=tol,
add_null=True)
# Final result should be padded to the correct size
self.assertTrue(len(v) == n_components)
self.assertTrue(U.shape[0] == n_components)
# All padded eigenvalues should be equal to 0
self.assertEqual(np.max(np.abs(v[n_eig:])), 0)
def test_no_add_null(self):
"""
Tests that warning is raised when n_components is
greater than the number of significant eigenpairs and add_null=False
"""
v, _ = np.linalg.eig(self.X)
tol = 1e-3
n_eig = len(v[v >= tol])
n_components = self.X.shape[0]
with warnings.catch_warnings(record=True) as w:
v, U = eig_solver(self.X,
n_components=n_components,
tol=tol,
add_null=False)
self.assertEqual(
str(w[-1].message),
f"There are fewer than {n_components} "
"significant eigenpair(s). Resulting decomposition"
f"will be truncated to {n_eig} eigenpairs.",
)
def test_reverse_sorted(self):
"""
Tests that the eigenvalues returned are in inverse order
"""
v, _ = eig_solver(self.X)
self.assertTrue(all([vv >= max(v[i:]) for i, vv in enumerate(v)]))