01 - Quantum Neural Networks

# This code is at:
#https://qiskit-community.github.io/qiskit-machine-learning/tutorials/01_neural_networks.html

from qiskit_machine_learning.utils import algorithm_globals

algorithm_globals.random_seed = 42

from qiskit.circuit import Parameter
from qiskit import QuantumCircuit

params1 = [Parameter("input1"), Parameter("weight1")]
qc1 = QuantumCircuit(1)
qc1.h(0)
qc1.ry(params1[0], 0)
qc1.rx(params1[1], 0)
qc1.draw("mpl", style="clifford")

from qiskit.quantum_info import SparsePauliOp

observable1 = SparsePauliOp.from_list([("Y" * qc1.num_qubits, 1)])

#from qiskit_machine_learning.neural_networks import EstimatorQNN
from quantumrings.toolkit.qiskit.machine_learning import QrEstimatorQNN as EstimatorQNN

estimator_qnn = EstimatorQNN(
    circuit=qc1, observables=observable1, input_params=[params1[0]], weight_params=[params1[1]]
)


estimator_qnn

<quantumrings.toolkit.qiskit.machine_learning.qr_estimator_qnn.QrEstimatorQNN at 0x236dd459450>

from qiskit.circuit import ParameterVector

inputs2 = ParameterVector("input", 2)
weights2 = ParameterVector("weight", 4)
print(f"input parameters: {[str(item) for item in inputs2.params]}")
print(f"weight parameters: {[str(item) for item in weights2.params]}")

qc2 = QuantumCircuit(2)
qc2.ry(inputs2[0], 0)
qc2.ry(inputs2[1], 1)
qc2.cx(0, 1)
qc2.ry(weights2[0], 0)
qc2.ry(weights2[1], 1)
qc2.cx(0, 1)
qc2.ry(weights2[2], 0)
qc2.ry(weights2[3], 1)

qc2.draw("mpl", style="clifford")

input parameters: ['input[0]', 'input[1]']
weight parameters: ['weight[0]', 'weight[1]', 'weight[2]', 'weight[3]']

#from qiskit_machine_learning.neural_networks import SamplerQNN
from quantumrings.toolkit.qiskit.machine_learning import QrSamplerQNN as SamplerQNN

sampler_qnn = SamplerQNN(circuit=qc2, input_params=inputs2, weight_params=weights2)
sampler_qnn

<quantumrings.toolkit.qiskit.machine_learning.qr_sampler_qnn.QrSamplerQNN at 0x236dfaefe90>

estimator_qnn_input = algorithm_globals.random.random(estimator_qnn.num_inputs)
estimator_qnn_weights = algorithm_globals.random.random(estimator_qnn.num_weights)

print(
    f"Number of input features for EstimatorQNN: {estimator_qnn.num_inputs} \nInput: {estimator_qnn_input}"
)
print(
    f"Number of trainable weights for EstimatorQNN: {estimator_qnn.num_weights} \nWeights: {estimator_qnn_weights}"
)

Number of input features for EstimatorQNN: 1
Input: [0.77395605]
Number of trainable weights for EstimatorQNN: 1
Weights: [0.43887844]

sampler_qnn_input = algorithm_globals.random.random(sampler_qnn.num_inputs)
sampler_qnn_weights = algorithm_globals.random.random(sampler_qnn.num_weights)

print(
    f"Number of input features for SamplerQNN: {sampler_qnn.num_inputs} \nInput: {sampler_qnn_input}"
)
print(
    f"Number of trainable weights for SamplerQNN: {sampler_qnn.num_weights} \nWeights: {sampler_qnn_weights}"
)

Number of input features for SamplerQNN: 2
Input: [0.85859792 0.69736803]
Number of trainable weights for SamplerQNN: 4
Weights: [0.09417735 0.97562235 0.7611397  0.78606431]

estimator_qnn_forward = estimator_qnn.forward(estimator_qnn_input, estimator_qnn_weights)

print(
    f"Forward pass result for EstimatorQNN: {estimator_qnn_forward}. \nShape: {estimator_qnn_forward.shape}"
)

Forward pass result for EstimatorQNN: [[0.2970094]].
Shape: (1, 1)

sampler_qnn_forward = sampler_qnn.forward(sampler_qnn_input, sampler_qnn_weights)

print(
    f"Forward pass result for SamplerQNN: {sampler_qnn_forward}.  \nShape: {sampler_qnn_forward.shape}"
)

Forward pass result for SamplerQNN: [[0.01269531 0.25878906 0.52734375 0.20117188]].
Shape: (1, 4)

estimator_qnn_forward_batched = estimator_qnn.forward(
    [estimator_qnn_input, estimator_qnn_input], estimator_qnn_weights
)

print(
    f"Forward pass result for EstimatorQNN: {estimator_qnn_forward_batched}.  \nShape: {estimator_qnn_forward_batched.shape}"
)

Forward pass result for EstimatorQNN: [[0.2970094]
 [0.2970094]].
Shape: (2, 1)

sampler_qnn_forward_batched = sampler_qnn.forward(
    [sampler_qnn_input, sampler_qnn_input], sampler_qnn_weights
)

print(
    f"Forward pass result for SamplerQNN: {sampler_qnn_forward_batched}.  \nShape: {sampler_qnn_forward_batched.shape}"
)

Forward pass result for SamplerQNN: [[0.01464844 0.26367188 0.51074219 0.2109375 ]
 [0.0234375  0.2578125  0.53613281 0.18261719]].
Shape: (2, 4)

estimator_qnn_input_grad, estimator_qnn_weight_grad = estimator_qnn.backward(
    estimator_qnn_input, estimator_qnn_weights
)

print(
    f"Input gradients for EstimatorQNN: {estimator_qnn_input_grad}.  \nShape: {estimator_qnn_input_grad}"
)
print(
    f"Weight gradients for EstimatorQNN: {estimator_qnn_weight_grad}.  \nShape: {estimator_qnn_weight_grad.shape}"
)

Input gradients for EstimatorQNN: None.
Shape: None
Weight gradients for EstimatorQNN: [[[0.63272767]]].
Shape: (1, 1, 1)

sampler_qnn_input_grad, sampler_qnn_weight_grad = sampler_qnn.backward(
    sampler_qnn_input, sampler_qnn_weights
)

print(
    f"Input gradients for SamplerQNN: {sampler_qnn_input_grad}.  \nShape: {sampler_qnn_input_grad}"
)
print(
    f"Weight gradients for SamplerQNN: {sampler_qnn_weight_grad}.  \nShape: {sampler_qnn_weight_grad.shape}"
)

Input gradients for SamplerQNN: None.
Shape: None
Weight gradients for SamplerQNN: [[[ 0.00585938 -0.12207031 -0.07861328 -0.10351562]
  [ 0.20458984 -0.08740234  0.07128906 -0.22460938]
  [-0.49267578  0.32617188 -0.31640625  0.10888672]
  [ 0.28222656 -0.11669922  0.32373047  0.21923828]]].
Shape: (1, 4, 4)

estimator_qnn.input_gradients = True
sampler_qnn.input_gradients = True

estimator_qnn_input_grad, estimator_qnn_weight_grad = estimator_qnn.backward(
    estimator_qnn_input, estimator_qnn_weights
)

print(
    f"Input gradients for EstimatorQNN: {estimator_qnn_input_grad}.  \nShape: {estimator_qnn_input_grad.shape}"
)
print(
    f"Weight gradients for EstimatorQNN: {estimator_qnn_weight_grad}.  \nShape: {estimator_qnn_weight_grad.shape}"
)

Input gradients for EstimatorQNN: [[[0.3038852]]].
Shape: (1, 1, 1)
Weight gradients for EstimatorQNN: [[[0.63272767]]].
Shape: (1, 1, 1)

sampler_qnn_input_grad, sampler_qnn_weight_grad = sampler_qnn.backward(
    sampler_qnn_input, sampler_qnn_weights
)

print(
    f"Input gradients for SamplerQNN: {sampler_qnn_input_grad}.  \nShape: {sampler_qnn_input_grad.shape}"
)
print(
    f"Weight gradients for SamplerQNN: {sampler_qnn_weight_grad}.  \nShape: {sampler_qnn_weight_grad.shape}"
)

Input gradients for SamplerQNN: [[[-0.07763672 -0.11572266]
  [ 0.39306641 -0.19482422]
  [-0.33203125  0.09814453]
  [ 0.01660156  0.21240234]]].
Shape: (1, 4, 2)
Weight gradients for SamplerQNN: [[[ 0.0078125  -0.10498047 -0.06787109 -0.10449219]
  [ 0.21386719 -0.08837891  0.06640625 -0.22070312]
  [-0.48779297  0.31542969 -0.32128906  0.08935547]
  [ 0.26611328 -0.12207031  0.32275391  0.23583984]]].
Shape: (1, 4, 4)

observable2 = SparsePauliOp.from_list([("Z" * qc1.num_qubits, 1)])

estimator_qnn2 = EstimatorQNN(
    circuit=qc1,
    observables=[observable1, observable2],
    input_params=[params1[0]],
    weight_params=[params1[1]],
)

estimator_qnn_forward2 = estimator_qnn2.forward(estimator_qnn_input, estimator_qnn_weights)
estimator_qnn_input_grad2, estimator_qnn_weight_grad2 = estimator_qnn2.backward(
    estimator_qnn_input, estimator_qnn_weights
)

print(f"Forward output for EstimatorQNN1: {estimator_qnn_forward.shape}")
print(f"Forward output for EstimatorQNN2: {estimator_qnn_forward2.shape}")
print(f"Backward output for EstimatorQNN1: {estimator_qnn_weight_grad.shape}")
print(f"Backward output for EstimatorQNN2: {estimator_qnn_weight_grad2.shape}")

Forward output for EstimatorQNN1: (1, 1)
Forward output for EstimatorQNN2: (1, 2)
Backward output for EstimatorQNN1: (1, 1, 1)
Backward output for EstimatorQNN2: (1, 2, 1)

parity = lambda x: "{:b}".format(x).count("1") % 2
output_shape = 2  # parity = 0, 1

sampler_qnn2 = SamplerQNN(
    circuit=qc2,
    input_params=inputs2,
    weight_params=weights2,
    interpret=parity,
    output_shape=output_shape,
)

sampler_qnn_forward2 = sampler_qnn2.forward(sampler_qnn_input, sampler_qnn_weights)
sampler_qnn_input_grad2, sampler_qnn_weight_grad2 = sampler_qnn2.backward(
    sampler_qnn_input, sampler_qnn_weights
)

print(f"Forward output for SamplerQNN1: {sampler_qnn_forward.shape}")
print(f"Forward output for SamplerQNN2: {sampler_qnn_forward2.shape}")
print(f"Backward output for SamplerQNN1: {sampler_qnn_weight_grad.shape}")
print(f"Backward output for SamplerQNN2: {sampler_qnn_weight_grad2.shape}")

Forward output for SamplerQNN1: (1, 4)
Forward output for SamplerQNN2: (1, 2)
Backward output for SamplerQNN1: (1, 4, 4)
Backward output for SamplerQNN2: (1, 2, 4)