def __init__(self,
maxiter: int = 1000,
offset: float = 10,
initialEta: float = 1,
tol: float = 1e-6,
disp: bool = False,
momentum: float = 0.25) -> None:
"""
Args:
maxiter: Maximum number of iterations, each iteration evaluation gradient.
offset: Number of iterations before eta starts decreasing
tol: The convergence criteria that must be reached before stopping.
Optimization stops when: absolute(loss - previous_loss) < tol
disp: Set to True to display convergence messages.
momentum: Bias towards the previous gradient momentum in current update.
Must be within the bounds: [0,1)
"""
validate_range_exclusive_max('momentum', momentum, 0, 1)
super().__init__()
self._initialEta = initialEta
self._offset = offset
self._maxiter = maxiter
self._tol = tol if tol is not None else 1e-6
self._disp = disp
self._momentum_coeff = momentum
self._previous_loss = None
def __init__(self,
maxiter: int = 1000,
eta: float = 3.0,
tol: float = 1e-6,
disp: bool = False,
momentum: float = 0.25) -> None:
"""
Args:
maxiter: Maximum number of iterations, each iteration evaluation gradient.
eta: The coefficient of the gradient update. Increasing this value
results in larger step sizes: param = previous_param - eta * deriv
tol: The convergence criteria that must be reached before stopping.
Optimization stops when: absolute(loss - previous_loss) < tol
disp: Set to True to display convergence messages.
momentum: Bias towards the previous gradient momentum in current update.
Must be within the bounds: [0,1)
"""
validate_range_exclusive_max('momentum', momentum, 0, 1)
super().__init__()
self._eta = eta
self._maxiter = maxiter
self._tol = tol if tol is not None else 1e-6
self._disp = disp
self._momentum_coeff = momentum
self._previous_loss = None
def __init__(self,
maxiter: Union[int, List[int]] = 1000,
eta: Union[float, List[float]] = 1.0,
tol: float = 1e-6, # this is tol
disp: bool = False,
momentum: Union[float, List[float]] = 0.25,
param_tol: float = 1e-6,
averaging: int = 10) -> None:
"""
Performs Analytical Quantum Gradient Descent (AQGD) with Epochs.
Args:
maxiter: Maximum number of iterations (full gradient steps)
eta: The coefficient of the gradient update. Increasing this value
results in larger step sizes: param = previous_param - eta * deriv
tol: Tolerance for change in windowed average of objective values.
Convergence occurs when either objective tolerance is met OR parameter
tolerance is met.
disp: Set to True to display convergence messages.
momentum: Bias towards the previous gradient momentum in current
update. Must be within the bounds: [0,1)
param_tol: Tolerance for change in norm of parameters.
averaging: Length of window over which to average objective values for objective
convergence criterion
Raises:
AquaError: If the length of ``maxiter``, `momentum``, and ``eta`` is not the same.
"""
super().__init__()
if isinstance(maxiter, int):
maxiter = [maxiter]
if isinstance(eta, float):
eta = [eta]
if isinstance(momentum, float):
momentum = [momentum]
if len(maxiter) != len(eta) or len(maxiter) != len(momentum):
raise AquaError("AQGD input parameter length mismatch. Parameters `maxiter`, `eta`, "
"and `momentum` must have the same length.")
for m in momentum:
validate_range_exclusive_max('momentum', m, 0, 1)
self._eta = eta
self._maxiter = maxiter
self._momenta_coeff = momentum
self._param_tol = param_tol
self._tol = tol
self._averaging = averaging
if disp:
warnings.warn('The disp parameter is deprecated as of '
'0.8.0 and will be removed no sooner than 3 months after the release. '
'The information is now available if you enable INFO level logging.',
DeprecationWarning, stacklevel=2)
self._disp = disp
# state
self._avg_objval = None
self._prev_param = None
self._eval_count = 0 # function evaluations
self._prev_loss = [] # type: List[float]
self._prev_grad = [] # type: List[List[float]]
def test_validate_range(self):
""" validate range test """
test_value = 2.5
with self.assertRaises(ValueError):
validate_range('test_value', test_value, 0, 2)
with self.assertRaises(ValueError):
validate_range('test_value', test_value, 3, 4)
validate_range('test_value', test_value, 2.5, 3)
validate_range_exclusive('test_value', test_value, 0, 3)
with self.assertRaises(ValueError):
validate_range_exclusive('test_value', test_value, 0, 2.5)
validate_range_exclusive('test_value', test_value, 2.5, 3)
validate_range_exclusive_min('test_value', test_value, 0, 3)
with self.assertRaises(ValueError):
validate_range_exclusive_min('test_value', test_value, 2.5, 3)
validate_range_exclusive_min('test_value', test_value, 0, 2.5)
validate_range_exclusive_max('test_value', test_value, 2.5, 3)
with self.assertRaises(ValueError):
validate_range_exclusive_max('test_value', test_value, 0, 2.5)
validate_range_exclusive_max('test_value', test_value, 2.5, 3)