def test_expected_shapes(self): val = tf.zeros((2, 3, 4, 5)) u, s, vh, _ = decompositions.svd_decomposition(tf, val, 2) self.assertEqual(u.shape, (2, 3, 6)) self.assertEqual(s.shape, (6, )) self.assertAllClose(s, np.zeros(6)) self.assertEqual(vh.shape, (6, 4, 5))
def svd_decomposition(self, tensor: Tensor, split_axis: int, max_singular_values: Optional[int] = None, max_truncation_error: Optional[float] = None ) -> Tuple[Tensor, Tensor, Tensor, Tensor]: return decompositions.svd_decomposition( self.tf, tensor, split_axis, max_singular_values, max_truncation_error)
def test_max_truncation_error_relative(self): absolute = np.diag([2.0, 1.0, 0.2, 0.1]) relative = np.diag([2.0, 1.0, 0.2, 0.1]) max_truncation_err = 0.2 _, _, _, trunc_sv_absolute = decompositions.svd_decomposition( tf, absolute, 1, max_truncation_error=max_truncation_err, relative=False) _, _, _, trunc_sv_relative = decompositions.svd_decomposition( tf, relative, 1, max_truncation_error=max_truncation_err, relative=True) np.testing.assert_almost_equal(trunc_sv_absolute, [0.1]) np.testing.assert_almost_equal(trunc_sv_relative, [0.2, 0.1])
def test_max_truncation_error(self): random_matrix = np.random.rand(10, 10) unitary1, _, unitary2 = np.linalg.svd(random_matrix) singular_values = np.array(range(10)) val = unitary1.dot(np.diag(singular_values).dot(unitary2.T)) u, s, vh, trun = decompositions.svd_decomposition( tf, val, 1, max_truncation_error=math.sqrt(5.1)) self.assertEqual(u.shape, (10, 7)) self.assertEqual(s.shape, (7, )) self.assertAllClose(s, np.arange(9, 2, -1)) self.assertEqual(vh.shape, (7, 10)) self.assertAllClose(trun, np.arange(2, -1, -1))