class TestInitialStateZero(QiskitAquaTestCase):
def test_qubits_2_vector(self):
self.zero = Zero(2)
cct = self.zero.construct_circuit('vector')
np.testing.assert_array_equal(cct, [1.0, 0.0, 0.0, 0.0])
def test_qubits_5_vector(self):
self.zero = Zero(5)
cct = self.zero.construct_circuit('vector')
np.testing.assert_array_equal(cct, [
1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
0.0, 0.0, 0.0, 0.0, 0.0, 0.0
])
def test_qubits_2_circuit(self):
self.zero = Zero(2)
cct = self.zero.construct_circuit('circuit')
self.assertEqual(cct.qasm(),
'OPENQASM 2.0;\ninclude "qelib1.inc";\nqreg q[2];\n')
def test_qubits_5_circuit(self):
self.zero = Zero(5)
cct = self.zero.construct_circuit('circuit')
self.assertEqual(cct.qasm(),
'OPENQASM 2.0;\ninclude "qelib1.inc";\nqreg q[5];\n')
def test_qubits_5_circuit(self):
""" Qubits 5 circuit test """
zero = Zero(5)
cct = zero.construct_circuit('circuit')
# pylint: disable=no-member
self.assertEqual(cct.qasm(),
'OPENQASM 2.0;\ninclude "qelib1.inc";\nqreg q[5];\n')
def test_qubits_5_vector(self):
""" Qubits 5 vector test """
zero = Zero(5)
cct = zero.construct_circuit('vector')
np.testing.assert_array_equal(cct, [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0])
class ADAPTVQE(QuantumAlgorithm):
CONFIGURATION = {
'name': 'ADAPTVQE',
'description': 'ADAPT-VQE Algorithm',
}
def __init__(
self,
operator_pool: OperatorPool,
initial_state: Union[InitialState, None],
vqe_optimizer: Optimizer,
hamiltonian: BaseOperator,
max_iters: int = 10,
grad_tol: float = 1e-3,
max_evals_grouped=1,
aux_operators=None,
auto_conversion=True,
use_zero_initial_parameters=False
):
super().__init__()
self.operator_pool = copy.deepcopy(operator_pool)
if initial_state is None:
self.initial_state = Zero(num_qubits=operator_pool.num_qubits)
else:
self.initial_state = initial_state
self.vqe_optimizer = vqe_optimizer
self.hamiltonian = hamiltonian
self.max_iters = max_iters
self.grad_tol = grad_tol
self.max_evals_grouped = max_evals_grouped
self.aux_operators = aux_operators
self.auto_conversion = auto_conversion
self.use_zero_initial_parameters = use_zero_initial_parameters
self.parameters_per_step = 1
self._coms = None
self.first_step = False
self._current_max_grad = None
self._optimal_circuit = None
self._ret = {}
self.adapt_step_history = {
'gradient_list': [], # updated in _compute_gradients
'max_gradient': [], # updated in _ansatz_operator_list
'optimal_parameters': [], # updated in _run
'circuit': [], # updated in _run
'operators': [], # updated in _ansatz_operator_list
'vqe_ret': [], # updated in _run
'energy_history': [], # updated in _run
'Total Eval Time': 0, #added by WillB
'Total num evals': 0 #added by WillB
}
@property
def _new_param(self):
if self.use_zero_initial_parameters:
if self.first_step:
return [0.0]
return [0.0 for i in range(self.parameters_per_step)]
else:
if self.first_step:
return [np.random.uniform(-np.pi, +np.pi)]
return [np.random.uniform(-np.pi, +np.pi) for i in range(self.parameters_per_step)]
def _run(self) -> dict:
start_time = time.time() #Added by WillB
logger.info('Starting ADAPT step {} of maximum {}'.format(1, self.max_iters))
circuit = self.initial_state.construct_circuit(mode='circuit')
params = self._new_param
op_list = self._ansatz_operator_list(current_circuit=circuit, current_ops=[])
vqe = self._vqe_run(operator_list=op_list, initial_parameters=params)
self.adapt_step_history['optimal_parameters'].append(vqe['optimal_params'])
self.adapt_step_history['circuit'].append(vqe['optimal_circuit'])
self.adapt_step_history['vqe_ret'].append(vqe['_ret'])
self.adapt_step_history['Total num evals'] += vqe['_ret']['num_optimizer_evals'] #added by WillB
self.adapt_step_history['energy_history'].append(vqe['_ret']['energy'])
logger.info('Finished ADAPT step {} of maximum {} with energy {}'.format(1, self.max_iters, vqe['_ret']['energy']))
iters = 1
print(iters)
while iters <= self.max_iters: #and self._current_max_grad > self.grad_tol: #removed by WillB
logger.info('Starting ADAPT step {} of maximum {}'.format(iters, self.max_iters))
circuit = vqe['optimal_circuit']
params = np.concatenate((vqe['optimal_params'], self._new_param))
op_list = self._ansatz_operator_list(current_circuit=circuit, current_ops=op_list)
#if self._current_max_grad < self.grad_tol: #removed by willB
# break
vqe = self._vqe_run(operator_list=op_list, initial_parameters=params)
self.adapt_step_history['optimal_parameters'].append(vqe['optimal_params'])
self.adapt_step_history['circuit'].append(vqe['optimal_circuit'])
self.adapt_step_history['vqe_ret'].append(vqe['_ret'])
self.adapt_step_history['Total num evals'] += vqe['_ret']['num_optimizer_evals'] #added by WillB
self.adapt_step_history['energy_history'].append(vqe['_ret']['energy'])
logger.info(
'Finished ADAPT step {} of maximum {} with energy {}'.format(iters, self.max_iters,
vqe['_ret']['energy']))
iters += 1
print(iters)
logger.info('Finished final ADAPT step {} of maximum {}, with final energy {}'.format(
iters,
self.max_iters,
vqe['_ret']['energy']
))
logger.info('Final gradient is {} where tolerance is {}'.format(
self._current_max_grad, self.grad_tol
))
self._optimal_circuit = circuit
self._ret = vqe['_ret']
del self.adapt_step_history['vqe_ret'] #added by WillB
stop_time = time.time() - start_time #added by WillB
self.adapt_step_history['Total Eval Time'] = stop_time #added by WillB
return self.adapt_step_history
def _compute_gradients(self, circuit: QuantumCircuit) -> List[Tuple[complex, complex]]:
kwargs = {'statevector_mode': self.quantum_instance.is_statevector}
logger.info('Constructing evaluation circuits...')
total_evaluation_circuits = list(parallel_map(
_circ_eval,
self.commutators,
task_kwargs={**kwargs, 'wave_function': circuit},
num_processes=aqua_globals.num_processes
))
total_evaluation_circuits = [item for sublist in total_evaluation_circuits for item in sublist]
logger.info('Removing duplicate circuits')
final_circs = []
for circ in total_evaluation_circuits:
if not fast_circuit_inclusion(circ, final_circs):
final_circs.append(circ)
logger.info('Finished removing duplicate circuits')
logger.debug('Executing {} circuits for gradient evaluation...'.format(len(final_circs)))
result = self.quantum_instance.execute(final_circs)
logger.debug('Computing {} gradients...'.format(len(self.commutators)))
grads = list(parallel_map(
_compute_grad,
self.commutators,
task_kwargs={**kwargs, 'result': result},
num_processes=aqua_globals.num_processes
))
self.adapt_step_history['Total num evals'] += len(final_circs)
logger.debug('Computed gradients: {}'.format(grads))
return [abs(tup[0]) for tup in grads]
@property
def commutators(self) -> List[BaseOperator]:
if self._coms is not None:
return self._coms
logger.info('Computing commutators...')
self._coms = list(parallel_map(
_commutator,
self.operator_pool.pool,
task_kwargs={'hamiltonian': self.hamiltonian, 'gradient_tolerance': self.grad_tol},
num_processes=aqua_globals.num_processes
)) # type: List[BaseOperator]
logger.info('Computed {} commutators'.format(len(self._coms)))
if all(isinstance(op, WeightedPauliOperator) for op in self._coms):
new_coms = []
new_pool = []
for com, op in zip(self._coms, self.operator_pool.pool):
if len(com.paulis) > 0:
new_coms.append(com)
new_pool.append(op)
self._coms = new_coms
self.operator_pool._pool = new_pool
logger.info('Dropped commuting terms, new pool has size {}'.format(len(self._coms)))
else:
logger.info(
'Dropping commuting terms currently only supported for WeightedPauliOperator class')
if len(self._coms) == 0:
raise ValueError('List of commutators is empty.')
return self._coms
def _ansatz_operator_list(self, current_circuit: QuantumCircuit,
current_ops: List[BaseOperator]) -> List[BaseOperator]:
grads = self._compute_gradients(circuit=current_circuit)
self._current_max_grad = np.max(grads)
self.adapt_step_history['max_gradient'].append(self._current_max_grad)
new_op = self.operator_pool.pool[np.argmax(grads)] # type: BaseOperator
logger.info('New operator with gradient {} added to ansatz: {}'.format(
self._current_max_grad,
str(new_op)
))
self.adapt_step_history['operators'].append(new_op.print_details())
return current_ops + [new_op]
def _vqe_run(self, operator_list: List[BaseOperator], initial_parameters: np.ndarray,
**kwargs) -> dict:
self._current_operator_list = operator_list
self._current_initial_parameters = initial_parameters
var_form = self.variational_form()
vqe = VQE(
operator=self.hamiltonian,
var_form=var_form,
optimizer=self.vqe_optimizer,
initial_point=initial_parameters,
max_evals_grouped=self.max_evals_grouped,
aux_operators=self.aux_operators,
callback=None,
auto_conversion=self.auto_conversion
)
vqe_result = vqe.run(self.quantum_instance, **kwargs) # == vqe._ret
return {
'optimal_circuit': vqe.get_optimal_circuit(),
'optimal_params': vqe.optimal_params,
'_ret': vqe_result
}
def variational_form(self):
return ADAPTVariationalForm(
operator_pool=self._current_operator_list,
bounds=[(-np.pi, +np.pi)] * len(self._current_operator_list),
initial_state=self.initial_state
)
def get_optimal_circuit(self):
if self._optimal_circuit is None:
raise AquaError(
"Cannot find optimal circuit before running the algorithm to find optimal params.")
return self._optimal_circuit
class ADAPTVQEROTO(ADAPTVQE):
def __init__(
self,
operator_pool,
initial_state,
vqe_optimizer,
hamiltonian,
max_iters,
energy_step_tol= 1e-8,
auto_conversion=True,
use_zero_initial_parameters=False,
postprocessing = False,
include_parameter = True
): #most of these params no longer matter, should edit to remove necessity
self._quantum_instance = None
super().__init__(operator_pool,
initial_state,
vqe_optimizer,
hamiltonian,
max_iters, #mximum number of iterations
grad_tol = 1e-3, #stopping criteria if we were using gradient tolerance (we're not)
max_evals_grouped=1,#unsure what this means
aux_operators=None, #additional operators in pool
auto_conversion=True, #when using VQE algorithm auto converts operators to best representaiton for backend
use_zero_initial_parameters=False #Make your parameters start at zero before gradient based optimization
) #initialization from ADAPTVQE
self.energy_step_tol = energy_step_tol #the stopping criteria for energy step
self.hamiltonian = hamiltonian
self.postprocessing = postprocessing
self.include_parameter = include_parameter
if initial_state is None: #creates new initial state if none passed
self.initial_state = Zero(num_qubits=operator_pool.num_qubits) #Custom(hamiltonian.num_qubits, state='uniform')
else:
self.initial_state = initial_state
#clean up old dictionary
del self.adapt_step_history['gradient_list']
del self.adapt_step_history['max_gradient']
#del self.adapt_step_history['vqe_ret']
#self.adapt_step_history.update({'Total Eval Time': 0})
#self.adapt_step_history.update({'Total num Evals': 0})
self.adapt_step_history.update({'Total number energy iterations': 0})
self.adapt_step_history.update({'Max Energy Step': 0})
def find_optim_param_energy(self, preferred_op = None, preferred_op_mode = False, previous_circuit = None) -> dict:
"""method: find_optim_param_energy
finds the optimum operator+parameter pair for the next iteration of the ansatz
args:
preferred_op - an operator (weightedpaulioperator) predetermined to be the next operator in the ansatz
(essentially converts this method to rotosolve instead of rotoselect)
preferred_op_mode - a flag (boolean) to choose whether or not preferred op should be used
returns:
dictionary with: optimal operator, name of optimal operator (for output purposes), optimal parameter,
optimal energy, A, B, and C for Asin(theta + B) + C
"""
A = np.empty(0)
C = np.empty(0)
final_circs = []
E_list = []
#measure energies (theta = 0, theta = pi/4, theta = -pi/4)
if(self.adapt_step_history['Total number energy iterations'] == 0):
wavefunc = self.initial_state.construct_circuit()
Energy_0_circ = self.hamiltonian.construct_evaluation_circuit(wavefunc, True)
result = self.quantum_instance.execute(Energy_0_circ)
Energy_0 = np.real(self.hamiltonian.evaluate_with_result(result, True)[0])
self.adapt_step_history['Total num evals'] += 1
else:
op_list = self._current_operator_list
bounds=[(-np.pi, +np.pi)]*len(op_list)
vf = ADAPTVariationalForm(op_list, bounds, self.initial_state)
wavefunc = vf.construct_circuit(self.adapt_step_history['optimal_parameters'][-1])
Energy_0 = self.adapt_step_history['energy_history'][-1]
if preferred_op_mode:
pool = preferred_op
else:
pool = self.operator_pool.pool
args = []
kwargs = {'ham': self.hamiltonian, 'energy_step_tol': self.energy_step_tol, 'parameter': np.pi/4}
op_list_pi4 = list(parallel_map(
Generate_op,
pool,
args,
kwargs
))
kwargs['parameter'] = -np.pi/4
op_list_negpi4 = list(parallel_map(
Generate_op,
self.operator_pool.pool,
args,
kwargs
))
op_list = op_list_pi4 + op_list_negpi4
kwargs = {'statevector_mode': self.quantum_instance.is_statevector}
total_evaluation_circuits = list(parallel_map(
_circ_eval,
op_list,
task_kwargs={**kwargs, 'wave_function': wavefunc},
num_processes=aqua_globals.num_processes
))
total_evaluation_circuits = [item for sublist in total_evaluation_circuits for item in sublist]
logger.info('Removing duplicate circuits')
for circ in total_evaluation_circuits:
if not fast_circuit_inclusion(circ, final_circs):
final_circs.append(circ)
result = self.quantum_instance.execute(final_circs)
Energies = list(parallel_map(
_compute_energy,
op_list,
task_kwargs={**kwargs, 'result': result},
num_processes=aqua_globals.num_processes
))
for entry in Energies:
E_list.append(np.real(entry[0]))
cutoff = int(len(E_list)/2)
Energy_pi4 = np.array(E_list[0:cutoff])
Energy_negpi4 = np.array(E_list[cutoff:])
#calculate minimum energy + A,B, and C from measured energies
B = np.arctan2(( -Energy_negpi4 - Energy_pi4 + 2*Energy_0),(Energy_pi4 - Energy_negpi4))
Optim_param_array = (-B - np.pi/2)/2
X = np.sin(B)
Y = np.sin(B + np.pi/2)
Z = np.sin(B - np.pi/2)
for i in range(0,(len(Energy_negpi4)-1)):
if Y[i] != 0:
C = np.append(C, (Energy_0-Energy_pi4[i]*(X[i]/Y[i]))/(1-X[i]/Y[i]))
A = np.append(A, (Energy_pi4[i] - C[-1])/Y[i])
else:
C = np.append(C, Energy_pi4[i])
A = np.append(A, (Energy_0 - C[-1])/X[i])
Optim_energy_array = C - A
#find minimum energy index
Optim_param_pos = np.argmin(Optim_energy_array)
min_energy = Optim_energy_array[Optim_param_pos]
Optim_param = Optim_param_array[Optim_param_pos]
#CPU has limit on smallest number to be calculated - looks like its somewhere around 1e-16
#manually set this to zero as it should be zero.
if min_energy > Energy_0 and abs(Optim_param) < 2e-16:
Optim_param = 0
min_energy = Energy_0
#find optimum operator
Optim_operator = self.operator_pool.pool[Optim_param_pos]
Optim_operator_name = self.operator_pool.pool[Optim_param_pos].print_details()
#keep track of number of quantum evaluations
self.adapt_step_history['Total num evals'] += len(final_circs)
return {'Newly Minimized Energy': min_energy, 'Next Parameter value': Optim_param,
'Next Operator identity': Optim_operator, 'Next Operator Name': Optim_operator_name, 'A': A[Optim_param_pos], 'B': B[Optim_param_pos], 'C': C[Optim_param_pos]}
def recent_energy_step(self):
return abs(self.adapt_step_history['energy_history'][-1] - self.adapt_step_history['energy_history'][-2])
#Our new run function, this will take time to edit
def run(self, quantum_instance) -> dict: #we'll keep this structure of returning a dictionary, though maybe I'll return the full changelog,
start = time.time()
self._current_operator_list = []
self._quantum_instance = quantum_instance
while self.adapt_step_history['Total number energy iterations'] <= 1:
logger.info('Starting ADAPTROTO step {} of maximum {}'.format(self.adapt_step_history['Total number energy iterations'], self.max_iters))
New_minimizing_data = self.find_optim_param_energy()
if self.adapt_step_history['Total number energy iterations'] > 1: #and New_minimizing_data['Newly Minimized Energy'] > self.adapt_step_history['energy_history'][-1]:
break
self._current_operator_list.append(New_minimizing_data['Next Operator identity'])
if self.include_parameter == True:
new_parameter = float(New_minimizing_data['Next Parameter value'])
else:
new_parameter = 0
if self.adapt_step_history['Total number energy iterations'] == 0:
self.adapt_step_history['optimal_parameters'].append([new_parameter])
else:
new_list = self.adapt_step_history['optimal_parameters'][-1] + [new_parameter]
self.adapt_step_history['optimal_parameters'].append(new_list)
self.adapt_step_history['operators'].append(New_minimizing_data['Next Operator Name'])
if self.postprocessing:
vqe_rotosolve_result = self._vqe_run(self._current_operator_list,np.array(self.adapt_step_history['optimal_parameters'][-1]))
self.adapt_step_history['optimal_parameters'].append(list(vqe_rotosolve_result['optimal_params']))
self.adapt_step_history['energy_history'].append(vqe_rotosolve_result['_ret']['energy'])
self.adapt_step_history['Total num evals'] += vqe_rotosolve_result['_ret']['num_optimizer_evals']
else:
self.adapt_step_history['energy_history'].append(New_minimizing_data['Newly Minimized Energy'])
logger.info('Finished ADAPTROTO step {} of maximum {} with energy {}'.format(self.adapt_step_history['Total number energy iterations'], self.max_iters, self.adapt_step_history['energy_history'][-1]))
self.adapt_step_history['Total number energy iterations'] += 1
print(self.adapt_step_history['Total number energy iterations'])
if self.recent_energy_step() > self.adapt_step_history['Max Energy Step']:
self.adapt_step_history['Max Energy Step'] = self.recent_energy_step()
while self.adapt_step_history['Total number energy iterations'] <= self.max_iters: #and self.recent_energy_step() >= self.energy_step_tol:
logger.info('Starting ADAPTROTO step {} of maximum {}'.format(self.adapt_step_history['Total number energy iterations'], self.max_iters))
New_minimizing_data = self.find_optim_param_energy()
#if New_minimizing_data['Newly Minimized Energy'] > self.adapt_step_history['energy_history'][-1]:
# break
self._current_operator_list.append(New_minimizing_data['Next Operator identity'])
if self.include_parameter == True:
new_parameter = float(New_minimizing_data['Next Parameter value'])
else:
new_parameter = 0
self.adapt_step_history['optimal_parameters'].append(self.adapt_step_history['optimal_parameters'][-1] + [new_parameter])
self.adapt_step_history['operators'].append(New_minimizing_data['Next Operator Name'])
if self.postprocessing:
vqe_rotosolve_result = self._vqe_run(self._current_operator_list, np.array(self.adapt_step_history['optimal_parameters'][-1]))
self.adapt_step_history['optimal_parameters'].append(list(vqe_rotosolve_result['optimal_params']))
self.adapt_step_history['energy_history'].append(vqe_rotosolve_result['_ret']['energy'])
self.adapt_step_history['Total num evals'] += vqe_rotosolve_result['_ret']['num_optimizer_evals']
else:
self.adapt_step_history['energy_history'].append(New_minimizing_data['Newly Minimized Energy'])
logger.info('Finished ADAPTROTO step {} of maximum {} with energy {}'.format(self.adapt_step_history['Total number energy iterations'], self.max_iters, self.adapt_step_history['energy_history'][-1]))
self.adapt_step_history['Total number energy iterations'] += 1
print(self.adapt_step_history['Total number energy iterations'])
if self.recent_energy_step() > self.adapt_step_history['Max Energy Step']:
self.adapt_step_history['Max Energy Step'] = self.recent_energy_step()
logger.info('Final energy step is {} where tolerance is {}'.format(
self.recent_energy_step(), self.energy_step_tol
))
eval_time = time.time() - start
self.adapt_step_history['Total Eval Time'] = eval_time
return self.adapt_step_history #return final minimized energy list
def test_qubits_2_circuit(self):
""" Qubits 2 Circuit test """
zero = Zero(2)
cct = zero.construct_circuit('circuit')
self.assertEqual(cct.qasm(),
'OPENQASM 2.0;\ninclude "qelib1.inc";\nqreg q[2];\n')
