def mapToPhysicalQubits(op, ansatz, logical2PhysicalMap):
n_qubits = max(logical2PhysicalMap) + 1
ir = op.toXACCIR()
xacc.setOption('qubit-map',
','.join([str(i) for i in logical2PhysicalMap]))
irp = xacc.getIRPreprocessor('qubit-map-preprocessor')
irp.process(ir)
ham = PauliOperator()
ham.fromXACCIR(ir)
ansatzir = xacc.gate.createIR()
ansatzir.addKernel(ansatz)
irp.process(ansatzir)
xacc.unsetOption('qubit-map')
return ham, ansatz, n_qubits
# logging.getLogger().setLevel(logging.DEBUG)
# requests_log = logging.getLogger("requests.packages.urllib3")
# requests_log.setLevel(logging.DEBUG)
# requests_log.propagate = True
# requests.get('https://httpbin.org/headers')
# File names to pass in for training and testing data
train_file = os.path.join('nist_data/optdigits.tra')
test_file = os.path.join('nist_data/optdigits.tes')
# Initialize the xacc framework
xacc.Initialize()
# Get the XACC D-Wave accelerator
dwave = xacc.getAccelerator('dwave')
# Select which D-Wave solver to use
xacc.setOption('dwave-solver', 'DW_2000Q_VFYC_2_1')
# Create an AcceleratorBuffer for use in the XACC D-Wave accelerator
buffer = dwave.createBuffer("buffer")
# Add embedding to AcceleratorBuffer
f = open('embedding.txt', 'r')
embedding = ast.literal_eval(f.read())
f.close()
buffer.addExtraInfo('embedding', embedding)
# Specify the settings for the algorithm execution
# These can also be passed directly into the xacc.qpu() decorator as keyword arguments
# an optional argument for naming the output buffer file
# is 'output': 'name_of_file' (no file extension)
settings = {'algo': "mnist_digit_train",
'accelerator': dwave,
'train_data': train_file,
def __call__(self, *args, **kwargs):
super().__call__(*args, **kwargs)
self.initial_rbm = True
self.buffer = args[0]
self.embedding = self.buffer.getInformation('embedding')
# Get the training parameters (rate, num_epochs, batch_size, momentum)
self.learn_rate = self.kwargs['rate']
self.num_epochs = self.kwargs['num_epochs']
self.momentum = self.kwargs['momentum']
self.batch_size = self.kwargs['batch_size']
self.num_classes = self.kwargs['max_classes']
self.train_steps = self.kwargs['train_steps']
if 'chain-strength' in self.kwargs:
xacc.setOption('chain-strength', self.kwargs['chain_strength'])
if 'num_samples' in self.kwargs:
xacc.setOption('dwave-num-reads', self.kwargs['num_samples'])
# Get parameterized DWK
self.rbm_function = self.compiledKernel
# Might be a better way to get these values, but this is what I'm shooting for right now
self.numV = 0
self.numH = 0
self.numW = 0
for inst in self.rbm_function.getParameters():
if 'v' in inst:
self.numV += 1
if 'h' in inst:
self.numH += 1
if 'w' in inst:
self.numW += 1
# Initializing the weights from a random normal distribution
# Initializing the hidden and visible biases to be zero
self.weights = np.random.normal(0.01, 1.0, (self.numV, self.numH))
self.visible_bias = np.zeros((1, self.numV))
self.hidden_bias = np.zeros((1, self.numH))
tic = time.clock()
self.data, self.n_evts = self.readTrainData(self.kwargs['train_data'])
self.data = self.batchData(self.data, self.batch_size)
for epoch in range(self.num_epochs):
train_step = 0
for batch in self.data:
xacc.info("Train Step {}".format(train_step))
if train_step >= self.train_steps > -1:
break
# get data expectation values
dataExpW, dataExpV, dataExpH = self.getDataExpectations(batch)
# set the RBM
self.setRBM()
training_buffer = self.qpu.createBuffer("buffer")
training_buffer.addExtraInfo('embedding', self.embedding)
# Execute the RBM with the new parameters
self.executeRBM(training_buffer)
# Get the expectation values from the D-Wave execution
modelExpW, modelExpV, modelExpH = self.getExpectations()
# Update the parameters for this batch
self.updateParameters(dataExpW, modelExpW, dataExpV, modelExpV,
dataExpH, modelExpH)
train_step += 1
self.buffer.addExtraInfo("rbm_visible",
self.visible_bias.flatten().tolist())
self.buffer.addExtraInfo("rbm_hidden",
self.hidden_bias.flatten().tolist())
self.buffer.addExtraInfo("rbm_weights",
self.weights.flatten().tolist())
toc = time.clock()
training_time = toc - tic
if 'test_data' in self.kwargs:
self.test_data, self.test_targets, n_tests = self.readTestData(
self.kwargs['test_data'])
self.test_data = self.batchData(self.test_data, self.batch_size)
evals = np.zeros((n_tests, self.num_classes))
truth_vals = np.zeros(n_tests)
for digit in range(self.num_classes):
w = np.reshape(
np.array(self.buffer.getInformation("rbm_weights")),
(self.numV, self.numH))
v = np.reshape(
np.array(self.buffer.getInformation("rbm_visible")),
(1, self.numV))
h = np.reshape(
np.array(self.buffer.getInformation("rbm_hidden")),
(1, self.numH))
count = 0
for i, batch in enumerate(self.test_data):
Fv = self.freeEnergy(batch, w, v, h)
targets = np.reshape(self.test_targets[i], (-1, ))
ll = list(zip(targets, Fv))
for j, item in enumerate(ll):
idx = i * self.batch_size + j
evals[idx, digit] = item[1]
if digit == 0:
truth_vals[idx] = item[0]
softmax = np.exp(-evals) / sum(np.exp(-evals))
predictions = np.argmax(softmax, axis=1)
accuracy = np.sum(predictions == truth_vals) / len(truth_vals)
self.buffer.addExtraInfo('accuracy', str(accuracy))
timestr = time.strftime("%Y%m%d-%H%M%S")
self.buffer.addExtraInfo("training-time", training_time)
print("Finished Training")
print("Accuracy: ", self.buffer.getInformation('accuracy'))
print("Training Time: ", self.buffer.getInformation('training-time'))
output_name = self.kwargs[
'output'] + '.ab' if 'output' in self.kwargs else "trained-rbm-buffer-{}.ab".format(
timestr)
f = open(output_name, "w")
f.write(str(self.buffer))
f.close()
return
def execute(self, inputParams):
"""
This method is intended to be inherited by vqe and vqe_energy subclasses to allow algorithm-specific implementation.
This superclass method adds extra information to the buffer and allows XACC settings options to be set before executing VQE.
Parameters:
inputParams : dictionary
a dictionary of input parameters obtained from .ini file
return QPU Accelerator buffer
Options used (obtained from inputParams):
'qubit-map': map of logical qubits to physical qubits
'n-execs': number of sampler executions of measurements
'initial-parameters': list of initial parameters for the VQE algorithm
'restart-from-file': AcceleratorDecorator option to allow restart of VQE algorithm
'readout-error': AcceleratorDecorator option for readout-error mitigation
"""
m = xacc.HeterogeneousMap()
if 'shots' in inputParams:
m.insert('shots', int(inputParams['shots']))
if 'backend' in inputParams:
m.insert('backend', inputParams['backend'])
self.qpu = xacc.getAccelerator(inputParams['accelerator'], m)
xaccOp = self.hamiltonian_generators[
inputParams['hamiltonian-generator']].generate(inputParams)
self.ansatz = self.ansatz_generators[inputParams['name']].generate(
inputParams, xaccOp.nBits())
if 'qubit-map' in inputParams:
qubit_map = ast.literal_eval(inputParams['qubit-map'])
xaccOp, self.ansatz, n_qubits = xaccvqe.mapToPhysicalQubits(
xaccOp, self.ansatz, qubit_map)
else:
n_qubits = xaccOp.nBits()
self.op = xaccOp
self.n_qubits = n_qubits
# create buffer, add some extra info (hamiltonian, ansatz-qasm, python-ansatz-qasm)
self.buffer = xacc.qalloc(n_qubits)
self.buffer.addExtraInfo('hamiltonian', self.op.toString())
self.buffer.addExtraInfo(
'ansatz-qasm',
self.ansatz.toString().replace('\\n', '\\\\n'))
pycompiler = xacc.getCompiler('pyxasm')
# self.buffer.addExtraInfo('ansatz-qasm-py', '\n'.join(pycompiler.translate(self.ansatz).split('\n')[1:]))
# heres where we can set up the algorithm Parameters
# all versions of vqe require: Accelerator, Ansatz, Observable
# pure-vqe requires Optimizer
# energy calculation has optional Parameters - can be random
self.vqe_options_dict = {
'accelerator': self.qpu,
'ansatz': self.ansatz,
'observable': self.op
}
# get optimizers for VQE
# needs to check if optimizer is a python plugin
# if not, use nlopt (with options)
# so we pull 'optimizer-options' out if available
# Optimizer-options needs to be passed to BOTH XACC Core optimizers and to python plugins
# vqe_options_dict is used to initialize the algorithms
self.optimizer = None
self.optimizer_options = {}
if 'optimizer-options' in inputParams:
self.optimizer_options = ast.literal_eval(
inputParams['optimizer-options'])
# initial-parameters for optimizer (vqe)
# parameters for vqe-energy
if 'initial-parameters' in inputParams:
self.optimizer_options['initial-parameters'] = ast.literal_eval(
inputParams['initial-parameters'])
if 'parameters' in inputParams:
self.optimizer_options['parameters'] = ast.literal_eval(
inputParams['parameters'])
if 'nlopt-maxeval' in inputParams:
self.optimizer_options['nlopt-maxeval'] = int(
inputParams['nlopt-maxeval'])
# check to see if optimizer is a python plugin
# if it is, we do not put it in self.vqe_options_dict
# if it is not, it is put there
if 'optimizer' in inputParams:
if inputParams['optimizer'] in self.vqe_optimizers:
self.optimizer = self.vqe_optimizers[inputParams['optimizer']]
else:
self.optimizer = xacc.getOptimizer(inputParams['optimizer'],
self.optimizer_options)
else:
self.optimizer = xacc.getOptimizer('nlopt', self.optimizer_options)
# vqe.py then will check vqe_options_dict for optimizer; if it isn't there, run python optimizer
# and of course if it is, we run with XACC
self.buffer.addExtraInfo('accelerator', inputParams['accelerator'])
# need to make sure the AcceleratorDecorators work correctly
if 'n-execs' in inputParams:
xacc.setOption('sampler-n-execs', inputParams['n-execs'])
self.qpu = xacc.getAcceleratorDecorator('improved-sampling',
self.qpu)
if 'restart-from-file' in inputParams:
xacc.setOption('vqe-restart-file',
inputParams['restart-from-file'])
self.qpu = xacc.getAcceleratorDecorator('vqe-restart', self.qpu)
self.qpu.initialize()
if 'readout-error' in inputParams and inputParams['readout-error']:
self.qpu = xacc.getAcceleratorDecorator('ro-error', self.qpu)
if 'rdm-purification' in inputParams and inputParams[
'rdm-purification']:
print("setting RDM Purification")
self.qpu = xacc.getAcceleratorDecorator('rdm-purification',
self.qpu)
m = xacc.HeterogeneousMap()
m.insert('fermion-observable', self.op)
self.qpu.initialize(m)
self.vqe_options_dict = {
'optimizer': self.optimizer,
'accelerator': self.qpu,
'ansatz': self.ansatz,
'observable': self.op
}
xacc.setOptions(inputParams)
def analyze(self, buffer, inputParams):
"""
This method is also to be inherited by vqe and vqe_energy subclasses to allow for algorithm-specific implementation.
This superclass method always generates a .csv file with measured expectation values for each kernel and calculated energy of each iteration.
Parameters:
inputParams : dictionary
a dictionary of input parameters obtained from .ini file
buffer : XACC AcceleratorBuffer
AcceleratorBuffer containing VQE results to be analyzed
Options used (in inputParams):
'readout-error': generate .csv file with readout-error corrected expectation values and calculated energy for each kernel and iteration.
'richardson-extrapolation': run Richardson-Extrapolation on the resulting Accelerator buffer (generating 4 more .csv files of expectation values and energies)
'rich-extra-iter': the number of iterations of Richardson-Extrapolation
"""
ps = buffer.getAllUnique('parameters')
timestr = time.strftime("%Y%m%d-%H%M%S")
exp_csv_name = "%s_%s_%s_%s" % (
os.path.splitext(buffer.getInformation('file-name'))[0],
buffer.getInformation('accelerator'), "exp_val_z", timestr)
f = open(exp_csv_name + ".csv", 'w')
exp_columns = [
c.getInformation('kernel')
for c in buffer.getChildren('parameters', ps[0])
] + ['<E>']
f.write(str(exp_columns).replace('[', '').replace(']', '') + '\n')
for p in ps:
energy = 0.0
for c in buffer.getChildren('parameters', p):
exp = c.getInformation('exp-val-z')
energy += exp * c.getInformation(
'coefficient') if c.hasExtraInfoKey('coefficient') else 0.0
f.write(str(exp) + ',')
f.write(str(energy) + '\n')
f.close()
if 'readout-error' in inputParams:
ro_exp_csv_name = "%s_%s_%s_%s" % (
os.path.splitext(buffer.getInformation('file-name'))[0],
buffer.getInformation('accelerator'), "ro_fixed_exp_val_z",
timestr)
f = open(ro_exp_csv_name + '.csv', 'w')
f.write(str(exp_columns).replace('[', '').replace(']', '') + '\n')
for p in ps:
energy = 0.0
for c in buffer.getChildren('parameters', p):
exp = c.getInformation('ro-fixed-exp-val-z')
energy += exp * c.getInformation(
'coefficient') if c.hasExtraInfoKey(
'coefficient') else 0.0
f.write(str(exp) + ',')
f.write(str(energy) + '\n')
f.close()
if 'richardson-extrapolation' in inputParams and inputParams[
'richardson-extrapolation']:
from scipy.optimize import curve_fit
import numpy as np
angles = buffer.getInformation('vqe-angles')
qpu = self.vqe_options_dict['accelerator']
self.vqe_options_dict[
'accelerator'] = xacc.getAcceleratorDecorator(
'rich-extrap', qpu)
self.vqe_options_dict['task'] = 'compute-energy'
xaccOp = self.op
self.vqe_options_dict['vqe-params'] = ','.join(
[str(x) for x in angles])
fileNames = {
r: "%s_%s_%s_%s" %
(os.path.splitext(buffer.getInformation('file-name'))[0],
buffer.getInformation('accelerator'), 'rich_extrap_' + str(r),
timestr) + '.csv'
for r in [1, 3, 5, 7]
}
nRE_Execs = 2 if not 'rich-extrap-iter' in inputParams else int(
inputParams['rich-extrap-iter'])
if nRE_Execs < 2:
print(
'Richardson Extrapolation needs more than 1 execution. Setting to 2.'
)
nRE_execs = 2
for r in [1, 3, 5, 7]:
f = open(fileNames[r], 'w')
xacc.setOption('rich-extrap-r', r)
for i in range(nRE_Execs):
richardson_buffer = qpu.createBuffer('q', self.n_qubits)
results = xaccvqe.execute(xaccOp, richardson_buffer,
**self.vqe_options_dict)
ps = richardson_buffer.getAllUnique('parameters')
for p in ps:
f.write(str(p).replace('[', '').replace(']', ''))
energy = 0.0
for c in richardson_buffer.getChildren(
'parameters', p):
exp = c.getInformation(
'ro-fixed-exp-val-z') if c.hasExtraInfoKey(
'ro-fixed-exp-val-z'
) else c.getInformation('exp-val-z')
energy += exp * c.getInformation('coefficient')
f.write(',' + str(exp))
f.write(',' + str(energy) + '\n')
f.close()
nParams = len(ps[0])
columns = ['t{}'.format(i) for i in range(nParams)]
kernelNames = [
c.getInformation('kernel')
for c in buffer.getChildren('parameters', ps[0])
]
columns += kernelNames
columns.append('E')
dat = [
np.genfromtxt(fileNames[1], delimiter=',', names=columns),
np.genfromtxt(fileNames[3], delimiter=',', names=columns),
np.genfromtxt(fileNames[5], delimiter=',', names=columns),
np.genfromtxt(fileNames[7], delimiter=',', names=columns)
]
allExps = [{k: [] for k in kernelNames} for i in range(4)]
allEnergies = []
temp = {r: [] for r in range(4)}
for i in range(nRE_Execs):
for r in range(4):
for term in kernelNames:
allExps[r][term].append(dat[r][term][i])
temp[r].append(dat[r]['E'][i])
evars = [np.std(temp[r]) for r in range(4)]
xVals = [1, 3, 5, 7]
avgExps = {
k: [np.mean(allExps[r][k]) for r in range(4)]
for k in kernelNames
}
varExps = {
k: [np.std(allExps[r][k]) for r in range(4)]
for k in kernelNames
}
energies = [np.mean(temp[r]) for r in range(4)]
def linear(x, a, b):
return a * x + b
def exp(x, a, b):
return a * np.exp(b * x) # + b
def quad(x, a, b, c):
return a * x * x + b * x + c
print('\nnoisy energy: ', energies[0], '+-', evars[0])
res = curve_fit(linear,
xVals,
energies, [1, energies[0]],
sigma=evars)
print('\nrich linear extrap: ', res[0][1], '+- ',
np.sqrt(np.diag(res[1])[1]))
res_exp = curve_fit(exp, xVals, energies, [0, 0], sigma=evars)
print('\nrich exp extrap: ', exp(0, res_exp[0][0], res_exp[0][1]),
'+-', np.sqrt(np.diag(res_exp[1])[1]))
res_q = curve_fit(quad, xVals, energies, [0, 0, 0], sigma=evars)
print("\nrich quad extrap: ",
quad(0, res_q[0][0], res_q[0][1], res_q[0][2]), "+-",
np.sqrt(np.diag(res_q[1])[2]))
def execute(self, inputParams):
"""
This method is intended to be inherited by vqe and vqe_energy subclasses to allow algorithm-specific implementation.
This superclass method adds extra information to the buffer and allows XACC settings options to be set before executing VQE.
Parameters:
inputParams : dictionary
a dictionary of input parameters obtained from .ini file
return QPU Accelerator buffer
Options used (obtained from inputParams):
'qubit-map': map of logical qubits to physical qubits
'n-execs': number of sampler executions of measurements
'initial-parameters': list of initial parameters for the VQE algorithm
'restart-from-file': AcceleratorDecorator option to allow restart of VQE algorithm
'readout-error': AcceleratorDecorator option for readout-error mitigation
"""
self.qpu = xacc.getAccelerator(inputParams['accelerator'])
xaccOp = self.hamiltonian_generators[
inputParams['hamiltonian-generator']].generate(inputParams)
self.ansatz = self.ansatz_generators[inputParams['name']].generate(
inputParams, xaccOp.nQubits())
if 'qubit-map' in inputParams:
qubit_map = ast.literal_eval(inputParams['qubit-map'])
xaccOp, self.ansatz, n_qubits = xaccvqe.mapToPhysicalQubits(
xaccOp, self.ansatz, qubit_map)
else:
n_qubits = xaccOp.nQubits()
self.op = xaccOp
self.n_qubits = n_qubits
self.buffer = self.qpu.createBuffer('q', n_qubits)
self.buffer.addExtraInfo('hamiltonian', str(xaccOp))
self.buffer.addExtraInfo(
'ansatz-qasm',
self.ansatz.toString('q').replace('\\n', '\\\\n'))
pycompiler = xacc.getCompiler('xacc-py')
self.buffer.addExtraInfo(
'ansatz-qasm-py',
'\n'.join(pycompiler.translate('q', self.ansatz).split('\n')[1:]))
self.optimizer = None
self.optimizer_options = {}
if 'optimizer' in inputParams:
if inputParams['optimizer'] in self.vqe_optimizers:
self.optimizer = self.vqe_optimizers[inputParams['optimizer']]
if 'method' in inputParams:
self.optimizer_options['method'] = inputParams['method']
if 'options' in inputParams:
self.optimizer_options['options'] = ast.literal_eval(
inputParams['options'])
if 'user-params' in inputParams:
self.optimizer_options['options'][
'user_params'] = ast.literal_eval(
inputParams['user-params'])
else:
xacc.setOption('vqe-backend', inputParams['optimizer'])
else:
xacc.info(
"No classical optimizer specified. Setting to default XACC optimizer."
)
self.buffer.addExtraInfo('accelerator', inputParams['accelerator'])
if 'n-execs' in inputParams:
xacc.setOption('sampler-n-execs', inputParams['n-execs'])
self.qpu = xacc.getAcceleratorDecorator('improved-sampling',
self.qpu)
if 'restart-from-file' in inputParams:
xacc.setOption('vqe-restart-file',
inputParams['restart-from-file'])
self.qpu = xacc.getAcceleratorDecorator('vqe-restart', self.qpu)
self.qpu.initialize()
if 'readout-error' in inputParams and inputParams['readout-error']:
self.qpu = xacc.getAcceleratorDecorator('ro-error', self.qpu)
if 'rdm-purification' in inputParams and inputParams[
'rdm-purification']:
self.qpu = xacc.getAcceleratorDecorator('rdm-purification',
self.qpu)
self.vqe_options_dict = {
'accelerator': self.qpu,
'ansatz': self.ansatz
}
if 'initial-parameters' in inputParams:
self.vqe_options_dict['vqe-params'] = ','.join([
str(x)
for x in ast.literal_eval(inputParams['initial-parameters'])
])
xacc.setOptions(inputParams)
from xaccvqe import PauliOperator
xacc.Initialize(['--compiler', 'quil'])
#ibm = xacc.getAccelerator('ibm')
tnqvm = xacc.getAccelerator('tnqvm')
buffer = tnqvm.createBuffer('q', 2)
ham = PauliOperator(5.906709445) + \
PauliOperator({0:'X',1:'X'}, -2.1433) + \
PauliOperator({0:'Y',1:'Y'}, -2.1433) + \
PauliOperator({0:'Z'}, .21829) + \
PauliOperator({1:'Z'}, -6.125)
xacc.setOption('ibm-shots', '8192')
#xacc.setOption('vqe-backend','vqe-bayesopt')
#xacc.setOption('bo-n-iter','20')
@xaccvqe.qpu.vqe(accelerator=tnqvm, observable=ham)
def ansatz(buffer, t0):
X(0)
Ry(t0, 1)
CNOT(1, 0)
# Run VQE with given ansatz kernel
initAngle = .5
ansatz(buffer, initAngle)
import xaccvqe
from xaccvqe import PauliOperator
import numpy as np
xacc.Initialize()
tnqvm = xacc.getAccelerator('tnqvm')
buffer = tnqvm.createBuffer('q', 2)
ham = PauliOperator(5.906709445) + \
PauliOperator({0:'X',1:'X'}, -2.1433) + \
PauliOperator({0:'Y',1:'Y'}, -2.1433) + \
PauliOperator({0:'Z'}, .21829) + \
PauliOperator({1:'Z'}, -6.125)
xacc.setOption('itensor-svd-cutoff', '1e-16')
# Hardware Efficient Ansatz xacc kernel
@xaccvqe.qpu.energy(accelerator=tnqvm, observable=ham)
def ansatz(buffer, *args):
xacc(hwe, layers=2, n_qubits=2, connectivity='[[0,1]]')
# XACC Kernels can display number of required nParameters
# and be persisted to qasm string
print(ansatz.nParameters())
print(ansatz.getFunction().toString('q'))
# Generate an initial random set of vqe params
# of the correct number of parameters and execute