def test_circuit_set_parameters_ungates(backend, accelerators, trainable):
"""Check updating parameters of circuit with list."""
original_backend = qibo.get_backend()
qibo.set_backend(backend)
params = [0.1, 0.2, 0.3, (0.4, 0.5), (0.6, 0.7, 0.8)]
if trainable:
trainable_params = list(params)
else:
trainable_params = [0.1, 0.3, (0.4, 0.5)]
c = Circuit(3, accelerators)
c.add(gates.RX(0, theta=0))
if trainable:
c.add(gates.CRY(0, 1, theta=0, trainable=trainable))
else:
c.add(gates.CRY(0, 1, theta=params[1], trainable=trainable))
c.add(gates.CZ(1, 2))
c.add(gates.U1(2, theta=0))
c.add(gates.CU2(0, 2, phi=0, lam=0))
if trainable:
c.add(gates.U3(1, theta=0, phi=0, lam=0, trainable=trainable))
else:
c.add(gates.U3(1, *params[4], trainable=trainable))
# execute once
final_state = c()
target_c = Circuit(3)
target_c.add(gates.RX(0, theta=params[0]))
target_c.add(gates.CRY(0, 1, theta=params[1]))
target_c.add(gates.CZ(1, 2))
target_c.add(gates.U1(2, theta=params[2]))
target_c.add(gates.CU2(0, 2, *params[3]))
target_c.add(gates.U3(1, *params[4]))
c.set_parameters(trainable_params)
np.testing.assert_allclose(c(), target_c())
# Attempt using a flat list
npparams = np.random.random(8)
if trainable:
trainable_params = np.copy(npparams)
else:
npparams[1] = params[1]
npparams[5:] = params[4]
trainable_params = np.delete(npparams, [1, 5, 6, 7])
target_c = Circuit(3)
target_c.add(gates.RX(0, theta=npparams[0]))
target_c.add(gates.CRY(0, 1, theta=npparams[1]))
target_c.add(gates.CZ(1, 2))
target_c.add(gates.U1(2, theta=npparams[2]))
target_c.add(gates.CU2(0, 2, *npparams[3:5]))
target_c.add(gates.U3(1, *npparams[5:]))
c.set_parameters(trainable_params)
np.testing.assert_allclose(c(), target_c())
qibo.set_backend(original_backend)
def test_circuit_set_parameters_ungates(backend, accelerators):
"""Check updating parameters of circuit with list."""
original_backend = qibo.get_backend()
qibo.set_backend(backend)
c = Circuit(3, accelerators)
c.add(gates.RX(0, theta=0))
c.add(gates.CRY(0, 1, theta=0))
c.add(gates.CZ(1, 2))
c.add(gates.U1(2, theta=0))
c.add(gates.CU2(0, 2, phi=0, lam=0))
c.add(gates.U3(1, theta=0, phi=0, lam=0))
# execute once
final_state = c()
params = [0.1, 0.2, 0.3, (0.4, 0.5), (0.6, 0.7, 0.8)]
target_c = Circuit(3)
target_c.add(gates.RX(0, theta=params[0]))
target_c.add(gates.CRY(0, 1, theta=params[1]))
target_c.add(gates.CZ(1, 2))
target_c.add(gates.U1(2, theta=params[2]))
target_c.add(gates.CU2(0, 2, *params[3]))
target_c.add(gates.U3(1, *params[4]))
c.set_parameters(params)
np.testing.assert_allclose(c(), target_c())
# Attempt using a flat list
params = np.random.random(8)
target_c = Circuit(3)
target_c.add(gates.RX(0, theta=params[0]))
target_c.add(gates.CRY(0, 1, theta=params[1]))
target_c.add(gates.CZ(1, 2))
target_c.add(gates.U1(2, theta=params[2]))
target_c.add(gates.CU2(0, 2, *params[3:5]))
target_c.add(gates.U3(1, *params[5:]))
c.set_parameters(params)
np.testing.assert_allclose(c(), target_c())
qibo.set_backend(original_backend)
def test_from_qasm_crotations():
import numpy as np
target = """OPENQASM 2.0;
qreg q[2];
crx(0.1) q[0],q[1];
crz(0.3) q[1],q[0];
cry(0.2) q[0],q[1];"""
c = Circuit.from_qasm(target)
assert c.depth == 3
assert isinstance(c.queue[0], gates.CRX)
assert isinstance(c.queue[1], gates.CRZ)
assert isinstance(c.queue[2], gates.CRY)
c2 = Circuit(2)
c2.add([gates.CRX(0, 1, 0.1), gates.CRZ(1, 0, 0.3), gates.CRY(0, 1, 0.2)])
np.testing.assert_allclose(c2().numpy(), c().numpy())
def fit(self,
reference,
initial_params=None,
batch_samples=128,
n_epochs=20000,
lr=0.5,
save=True):
"""Execute qGAN training.
Args:
reference (array): samples from the reference input distribution.
initial_parameters (array): initial parameters for the quantum generator. If not provided,
the default initial parameters will be used.
discriminator (:class:`tensorflow.keras.models`): custom classical discriminator. If not provided,
the default classical discriminator will be used.
batch_samples (int): number of training examples utilized in one iteration.
n_epochs (int): number of training iterations.
lr (float): initial learning rate for the quantum generator.
It controls how much to change the model each time the weights are updated.
save (bool): If ``True`` the results of training (trained parameters and losses)
will be saved on disk. Default is ``True``.
"""
if initial_params is None and self.circuit is not None:
raise_error(
ValueError,
"Set the initial parameters for your custom quantum generator."
)
elif initial_params is not None and self.circuit is None:
raise_error(
ValueError,
"Define the custom quantum generator to use custom initial parameters."
)
self.reference = reference
self.nqubits = reference.shape[1]
self.training_samples = reference.shape[0]
self.initial_params = initial_params
self.batch_samples = batch_samples
self.n_epochs = n_epochs
self.lr = lr
# create classical discriminator
if self.discriminator is None:
discriminator = self.define_discriminator()
else:
discriminator = self.discriminator
if discriminator.input_shape[1] is not self.nqubits:
raise_error(
ValueError,
"The number of input neurons in the discriminator has to be equal to "
"the number of qubits in the circuit (dimension of the input reference distribution)."
)
# create quantum generator
if self.circuit is None:
circuit = models.Circuit(self.nqubits)
for l in range(self.layers):
for q in range(self.nqubits):
circuit.add(gates.RY(q, 0))
circuit.add(gates.RZ(q, 0))
circuit.add(gates.RY(q, 0))
circuit.add(gates.RZ(q, 0))
for i in range(0, self.nqubits - 1):
circuit.add(gates.CRY(i, i + 1, 0))
circuit.add(gates.CRY(self.nqubits - 1, 0, 0))
for q in range(self.nqubits):
circuit.add(gates.RY(q, 0))
else:
circuit = self.circuit
if circuit.nqubits != self.nqubits:
raise_error(
ValueError,
"The number of qubits in the circuit has to be equal to "
"the number of dimensions in the reference distribution.")
# define hamiltonian to generate fake samples
def hamiltonian(nqubits, position):
identity = [[1, 0], [0, 1]]
m0 = hamiltonians.Z(1).matrix
kron = []
for i in range(nqubits):
if i == position:
kron.append(m0)
else:
kron.append(identity)
for i in range(nqubits - 1):
if i == 0:
ham = np.kron(kron[i + 1], kron[i])
else:
ham = np.kron(kron[i + 1], ham)
ham = hamiltonians.Hamiltonian(nqubits, ham)
return ham
hamiltonians_list = []
for i in range(self.nqubits):
hamiltonians_list.append(hamiltonian(self.nqubits, i))
# train model
self.train(discriminator, circuit, hamiltonians_list, save)