# This code is part of a Qiskit project.
#
# (C) Copyright IBM 2021, 2023.
#
# This code is licensed under the Apache License, Version 2.0. You may
# obtain a copy of this license in the LICENSE.txt file in the root directory
# of this source tree or at http://www.apache.org/licenses/LICENSE-2.0.
#
# Any modifications or derivative works of this code must retain this
# copyright notice, and modified files need to carry a notice indicating
# that they have been altered from the originals.
"""Implementation of the warm start QAOA optimizer."""
import copy
from abc import ABC, abstractmethod
from typing import Dict, List, Optional, Tuple, Union, cast
import numpy as np
from qiskit import QuantumCircuit
from qiskit.circuit import Parameter
from qiskit_algorithms import QAOA
from ..converters.quadratic_program_converter import QuadraticProgramConverter
from ..exceptions import QiskitOptimizationError
from ..problems.quadratic_program import QuadraticProgram
from ..problems.variable import VarType
from .minimum_eigen_optimizer import MinimumEigenOptimizationResult, MinimumEigenOptimizer
from .optimization_algorithm import OptimizationAlgorithm, OptimizationResultStatus, SolutionSample
[ドキュメント]class BaseAggregator(ABC):
"""A base abstract class for aggregates results"""
[ドキュメント] @abstractmethod
def aggregate(self, results: List[MinimumEigenOptimizationResult]) -> List[SolutionSample]:
"""
Aggregates the results.
Args:
results: List of result objects that need to be combined.
Returns:
Aggregated samples.
"""
raise NotImplementedError
[ドキュメント]class MeanAggregator(BaseAggregator):
"""Aggregates the results by averaging the probability of each sample."""
[ドキュメント] def aggregate(self, results: List[MinimumEigenOptimizationResult]) -> List[SolutionSample]:
"""
Args:
results: List of result objects that need to be combined.
Returns:
Aggregated samples by averaging them.
"""
# Use a dict for fast solution look-up
# Key: sample code, value: tuple of fval, probability
dict_samples: Dict[str, Tuple[float, float]] = {}
def _to_string(x: np.ndarray) -> str:
return "".join(str(int(v)) for v in x)
def _from_string(string) -> np.ndarray:
return np.array([float(c) for c in string])
# Sum up all the probabilities in the results
for result in results:
for sample in result.samples:
# state, fval, prob = sample[0], sample[1], sample[2]
state, fval, prob = (
_to_string(sample.x),
sample.fval,
sample.probability,
)
if state in dict_samples:
dict_samples[state] = (fval, dict_samples[state][1] + prob)
else:
dict_samples[state] = (fval, prob)
# Divide by the number of results to normalize
aggregated_samples = []
num_results = len(results)
for state, val_prob in dict_samples.items():
# sample = (state, dict_samples[state][0], dict_samples[state][1] / num_results)
sample = SolutionSample(
x=_from_string(state),
fval=val_prob[0],
probability=val_prob[1] / num_results,
status=OptimizationResultStatus.SUCCESS,
)
aggregated_samples.append(sample)
return aggregated_samples
[ドキュメント]class WarmStartQAOAFactory:
"""
A factory that produces quantum circuits for the QAOA implementation. The methods of this
factory can be overridden to modify behavior of QAOA. This implementation generates quantum
circuits for initial state and mixer to warm start QAOA.
"""
def __init__(self, epsilon: float) -> None:
"""
Args:
epsilon: the regularization parameter that changes the initial variables according to
xi = epsilon if xi < epsilon
xi = 1-epsilon if xi > epsilon.
The regularization parameter epsilon should be between 0 and 0.5. When it
is 0.5 then warm start corresponds to standard QAOA.
Raises:
QiskitOptimizationError: if ``epsilon`` is not in the range [0, 0.5].
"""
if epsilon < 0.0 or epsilon > 0.5:
raise QiskitOptimizationError(
f"Epsilon for warm-start QAOA needs to be between 0 and 0.5, "
f"actual value: {epsilon}"
)
self._epsilon = epsilon
[ドキュメント] def create_initial_variables(self, solution: np.ndarray) -> List[float]:
"""
Creates initial variable values to warm start QAOA.
Args:
solution: a solution obtained for the relaxed problem.
Returns:
A list of initial variables constructed from a relaxed solution.
"""
initial_variables = []
for variable in solution:
if variable < self._epsilon:
initial_variables.append(self._epsilon)
elif variable > 1.0 - self._epsilon:
initial_variables.append(1.0 - self._epsilon)
else:
initial_variables.append(variable)
return initial_variables
[ドキュメント] def create_initial_state(self, initial_variables: List[float]) -> QuantumCircuit:
"""
Creates an initial state quantum circuit to warm start QAOA.
Args:
initial_variables: Already created initial variables.
Returns:
A quantum circuit that represents initial state.
"""
circuit = QuantumCircuit(len(initial_variables))
for index, relaxed_value in enumerate(initial_variables):
theta = 2 * np.arcsin(np.sqrt(relaxed_value))
circuit.ry(theta, index)
return circuit
[ドキュメント] def create_mixer(self, initial_variables: List[float]) -> QuantumCircuit:
"""
Creates an evolved mixer circuit as Ry(theta)Rz(-2beta)Ry(-theta).
Args:
initial_variables: Already created initial variables.
Returns:
A quantum circuit to be used as a mixer in QAOA.
"""
circuit = QuantumCircuit(len(initial_variables))
beta = Parameter("beta")
for index, relaxed_value in enumerate(initial_variables):
theta = 2 * np.arcsin(np.sqrt(relaxed_value))
circuit.ry(-theta, index)
circuit.rz(-2.0 * beta, index)
circuit.ry(theta, index)
return circuit
[ドキュメント]class WarmStartQAOAOptimizer(MinimumEigenOptimizer):
"""A meta-algorithm that uses a pre-solver to solve a relaxed version of the problem.
Users must implement their own pre solvers by inheriting from the base class.
References:
[1]: Daniel J. Egger et al., Warm-starting quantum optimization.
`arXiv:2009.10095 <https://arxiv.org/abs/2009.10095>`_
"""
def __init__(
self,
pre_solver: OptimizationAlgorithm,
relax_for_pre_solver: bool,
qaoa: QAOA,
epsilon: float = 0.25,
num_initial_solutions: int = 1,
warm_start_factory: Optional[WarmStartQAOAFactory] = None,
aggregator: Optional[BaseAggregator] = None,
penalty: Optional[float] = None,
converters: Optional[
Union[QuadraticProgramConverter, List[QuadraticProgramConverter]]
] = None,
) -> None:
"""Initializes the optimizer. For correct initialization either
``epsilon`` or ``circuit_factory`` must be passed. If only ``epsilon`` is specified
(either an explicit value or the default one), then an instance of
:class:`~qiskit.optimization.algorithms.WarmStartQAOACircuitFactory` is created.
If ``circuit_factory`` is specified then this instance is used in the implementation
and ``epsilon`` value is ignored.
Args:
pre_solver: An instance of an optimizer to solve the relaxed version of the problem.
relax_for_pre_solver: True if the problem must be relaxed to the continuous case
before passing it to the pre-solver.
qaoa: A QAOA instance to be used in the computations.
epsilon: the regularization parameter that changes the initial variables according to
xi = epsilon if xi < epsilon
xi = 1-epsilon if xi > epsilon.
The regularization parameter epsilon should be between 0 and 0.5. When it
is 0.5 then warm start corresponds to standard QAOA. If ``circuit_factory`` is
specified then this parameter is ignored. Default value is 0.25.
num_initial_solutions: An optional number of relaxed (continuous) solutions to use.
Default value is 1.
warm_start_factory: An optional instance of the factory to warm start QAOA.
This factory is used to create circuits for initial state and mixer.
If ``None`` is specified then a default one,
an instance of :class:`~qiskit.optimization.algorithms.WarmStartQAOACircuitFactory`
is created using the value of the ``epsilon`` parameter.
aggregator: Class that aggregates different results. This is used if the pre-solver
returns several initial states.
penalty: The penalty factor to be used, or ``None`` for applying a default logic.
converters: The converters to use for converting a problem into a different form.
By default, when None is specified, an internally created instance of
:class:`~qiskit.optimization.converters.QuadraticProgramToQubo` will be used.
Raises:
QiskitOptimizationError: if ``epsilon`` is not in the range [0, 0.5].
"""
if epsilon < 0.0 or epsilon > 0.5:
raise QiskitOptimizationError(
f"Epsilon for warm-start QAOA needs to be between 0 and 0.5, "
f"actual value: {epsilon}"
)
self._pre_solver = pre_solver
self._relax_for_pre_solver = relax_for_pre_solver
self._qaoa = qaoa
self._epsilon = epsilon
self._num_initial_solutions = num_initial_solutions
if warm_start_factory is None:
warm_start_factory = WarmStartQAOAFactory(epsilon)
self._warm_start_factory = warm_start_factory
if num_initial_solutions > 1 and aggregator is None:
aggregator = MeanAggregator()
self._aggregator = aggregator
super().__init__(qaoa, penalty, converters)
[ドキュメント] def solve(self, problem: QuadraticProgram) -> MinimumEigenOptimizationResult:
"""Tries to solves the given problem using the optimizer.
The pre-solver is run to warm-start the solver. Next, the optimizer is run
to try to solve the optimization problem.
Args:
problem: The problem to be solved.
Returns:
The result of the optimizer applied to the problem.
Raises:
QiskitOptimizationError: If problem not compatible or the presolver can't solve
a problem.
"""
message = self.get_compatibility_msg(problem)
if len(message) > 0:
raise QiskitOptimizationError(f"Incompatible problem: {message}")
# convert problem to minimization QUBO or another form if converters are specified
converted_problem = self._convert(problem, self._converters)
# if the pre-solver can't solve the problem then it should be relaxed.
if self._relax_for_pre_solver:
pre_solver_problem = self._relax_problem(converted_problem)
else:
pre_solver_problem = converted_problem
opt_result = self._pre_solver.solve(pre_solver_problem)
if opt_result.status != OptimizationResultStatus.SUCCESS:
raise QiskitOptimizationError(
f"Presolver returned status {opt_result.status}, " f"the problem can't be solved"
)
# we pick only a certain number of the pre-solved solutions.
num_pre_solutions = min(self._num_initial_solutions, len(opt_result.samples))
pre_solutions = opt_result.samples[:num_pre_solutions]
# construct operator and offset
operator, offset = converted_problem.to_ising()
results: List[MinimumEigenOptimizationResult] = []
for pre_solution in pre_solutions:
# Set the solver using the result of the pre-solver.
initial_variables = self._warm_start_factory.create_initial_variables(pre_solution.x)
self._qaoa.initial_state = self._warm_start_factory.create_initial_state(
initial_variables
)
self._qaoa.mixer = self._warm_start_factory.create_mixer(initial_variables)
# approximate ground state of operator using min eigen solver.
results.append(self._solve_internal(operator, offset, converted_problem, problem))
if len(results) == 1:
# there's no need to call _interpret, it is already done by MinimumEigenOptimizer
return results[0]
else:
samples = self._aggregator.aggregate(results)
samples.sort(key=lambda sample: problem.objective.sense.value * sample.fval)
# translate result back to the original variables
return cast(
MinimumEigenOptimizationResult,
self._interpret(
x=samples[0].x,
problem=problem,
converters=self._converters,
result_class=MinimumEigenOptimizationResult,
samples=samples,
),
)
@staticmethod
def _relax_problem(problem: QuadraticProgram) -> QuadraticProgram:
"""
Change all variables to continuous.
Args:
problem: Problem to relax.
Returns:
A copy of the original problem where all variables are continuous.
"""
relaxed_problem = copy.deepcopy(problem)
for variable in relaxed_problem.variables:
variable.vartype = VarType.CONTINUOUS
return relaxed_problem