# This work was supported by the EuroHPC PL infrastructure funded at the
# Smart Growth Operational Programme (2014-2020), Measure 4.2
# under the grant agreement no. POIR.04.02.00-00-D014/20-00
from typing import Callable, Any, Type
import pennylane as qml
from pennylane import numpy as np
from QHyper.optimizers.base import (
Optimizer, OptimizationResult, OptimizerError, OptimizationParameter)
QML_GRADIENT_DESCENT_OPTIMIZERS: dict[
str, Type[qml.GradientDescentOptimizer]] = {
'adam': qml.AdamOptimizer,
'adagrad': qml.AdagradOptimizer,
'rmsprop': qml.RMSPropOptimizer,
'momentum': qml.MomentumOptimizer,
'nesterov_momentum': qml.NesterovMomentumOptimizer,
'sgd': qml.GradientDescentOptimizer,
'qng': qml.QNGOptimizer,
}
[docs]
class QmlGradientDescent(Optimizer):
"""Gradient Descent Optimizer
This minimizer is a wrapper for gradient descent optimizers
provided by PennyLane.
This alogrithm requries the following parameters to be set:
- `init` initial values for each parameter
Attributes
----------
optimizer : qml.GradientDescentOptimizer
object of class GradientDescentOptimizer or inheriting from this class
steps : int, default 200
number of optimization steps
stepsize : float, default 0.005
stepsize for the optimizer
verbose : bool, default False
if set to True, additional information will be printed
"""
optimizer: qml.GradientDescentOptimizer
steps: int
stepsize: float
verbose: bool
def __init__(
self,
name: str = 'adam',
steps: int = 200,
stepsize: float = 0.005,
verbose: bool = False,
**kwargs: Any
) -> None:
"""
Parameters
----------
name : str, default 'adam'
name of the gradient descent optimizer provided by PennyLane
steps : int, default 200
number of optimization steps
stepsize : float, default 0.005
stepsize for the optimizer
verbose : bool, default False
if set to True, additional information will be printed
**kwargs : Any
Additional arguments that will be passed to the PennyLane
optimizer. More infomation can be found in the PennyLane
documentation.
"""
if name not in QML_GRADIENT_DESCENT_OPTIMIZERS:
raise ValueError(
f'Optimizer {name} not found. '
'Available optimizers: '
f'{list(QML_GRADIENT_DESCENT_OPTIMIZERS.keys())}'
)
self.optimizer = QML_GRADIENT_DESCENT_OPTIMIZERS[name](
stepsize=stepsize,
**kwargs
)
self.steps = steps
self.verbose = verbose
[docs]
def minimize(
self,
func: Callable[[list[float]], OptimizationResult],
init: OptimizationParameter
) -> OptimizationResult:
init.assert_init()
if isinstance(self.optimizer, qml.QNGOptimizer):
raise OptimizerError(
'QNG is not supported via optimizer, use qml_qaoa instead')
def wrapper(params: list[float]) -> float:
return func(list(params)).value
cost_history = []
best_result = float('inf')
best_params = np.array(init.init, requires_grad=True)
params = np.array(init.init, requires_grad=True)
if hasattr(self.optimizer, 'reset'):
self.optimizer.reset() # type: ignore
for i in range(self.steps):
params, cost = self.optimizer.step_and_cost(wrapper, params)
params = np.array(params, requires_grad=True)
if cost < best_result:
best_params = params
best_result = cost
cost_history.append(OptimizationResult(float(cost), params))
if self.verbose:
print(f'Step {i+1}/{self.steps}: {float(cost)}')
return OptimizationResult(best_result, best_params, [cost_history])
[docs]
def minimize_expval_func(
self, func: qml.QNode, init: OptimizationParameter
) -> OptimizationResult:
"""
Used in :py:class:`.QML_QAOA` to minimize the
expectation value function.
"""
cost_history = []
cost = float('inf')
params = np.array(init.init, requires_grad=True)
if hasattr(self.optimizer, 'reset'):
self.optimizer.reset() # type: ignore
for i in range(self.steps):
params, cost = self.optimizer.step_and_cost(func, params)
params = np.array(params, requires_grad=True)
cost_history.append(OptimizationResult(float(cost), params))
if self.verbose:
print(f'Step {i+1}/{self.steps}: {float(cost)}')
return OptimizationResult(cost, params, [cost_history])
