def _post_update_n_dimensions_9(self) -> bool:
# I != U1 != U2 != I, D1 != I != D2
self._U[0] = random_orthogonal_matrix(self._n_dimensions, self._rng)
self._U[1] = random_orthogonal_matrix(self._n_dimensions, self._rng)
self._d[0] = self._random_d()
self._d[1] = self._random_d()
raise NotImplementedError()
def _post_update_n_dimensions(self) -> None:
if self._rotate:
self._rotation_matrix_y = random_orthogonal_matrix(
self._n_dimensions, self._rng)
self._rotation_matrix_z = random_orthogonal_matrix(
self._n_dimensions, self._rng)
else:
self._rotation_matrix_y = np.eye(self._n_dimensions)
self._rotation_matrix_z = np.eye(self._n_dimensions)
self._scaler_inv = self._a * self._a * self._n_dimensions
def _post_update_n_dimensions(self) -> None:
self._rotation_matrix = (random_orthogonal_matrix(
self._n_dimensions, self._rng) if self._rotate else np.eye(
self._n_dimensions))
self._y_max = 1.0 / np.linalg.norm(self._rotation_matrix[0], np.inf)
self._constraints_handler = BoxConstraintsHandler(
self._n_dimensions, self._bounds)
def _post_update_n_dimensions_6(self) -> bool:
# U1 = I, U2 != I, D2 != I
for i in range(self._n_dimensions):
self._U[0, i, i] = 1.0
self._U[1] = random_orthogonal_matrix(self._n_dimensions, self._rng)
self._d[1] = self._random_d()
raise NotImplementedError()
def _post_update_n_dimensions(self) -> None:
if self._rotate:
self._rotation_matrix_y = random_orthogonal_matrix(
self._n_dimensions, self._rng
)
self._rotation_matrix_z = random_orthogonal_matrix(
self._n_dimensions, self._rng
)
else:
self._rotation_matrix_y = np.eye(self._n_dimensions)
self._rotation_matrix_z = np.eye(self._n_dimensions)
self._scaler_inv = self._a * self._a * self._n_dimensions
self._coefficients = self._a ** (
2.0 * np.arange(self._n_dimensions) / (self._n_dimensions - 1)
)
def _post_update_n_dimensions(self) -> None:
n = self._n_dimensions
# Update box constraints
lower_bounds = np.repeat(-5.0, n)
lower_bounds[0] = 0.0
upper_bounds = np.repeat(5.0, n)
upper_bounds[0] = 1.0
self._constraints_handler = BoxConstraintsHandler(
self._n_dimensions, (lower_bounds, upper_bounds))
# Create the random orthogonal (and restricted) rotation matrix
self._rotation_matrix = np.zeros((n, n))
self._rotation_matrix[0, 0] = 1.0
self._rotation_matrix[1:,
1:] = random_orthogonal_matrix(n - 1, self._rng)
def _post_update_n_dimensions(self) -> None:
self._scaler_inv = self._a * self._a * self._n_dimensions
# The O matrix in the paper
orthogonal_matrix = random_orthogonal_matrix(
self._n_dimensions, self._rng
)
# The D matrix in the paper
diagonal_matrix = np.zeros((self._n_dimensions, self._n_dimensions))
for i in range(self._n_dimensions):
diagonal_matrix[i, i] = self._a ** (i / (self._n_dimensions - 1))
# The DO matrix in the paper
self._rotation_matrix = diagonal_matrix @ orthogonal_matrix
# The M matrix in the paper
self._create_centers_matrix()
def setUp(self) -> None:
"""Set up for tests."""
rng = np.random.default_rng(seed=0)
self.Q = random_orthogonal_matrix(100, rng)