Machine Learning API Reference¶
Quantum Kernels¶
kernels
¶
Quantum kernels (ZZ-feature-map, Havlicek et al. Nature 567, 2019).
Implements the ZZ feature map circuit and a quantum kernel SVM classifier for financial data classification tasks such as credit scoring and fraud detection.
References¶
Havlicek et al., Nature 567, 209-212 (2019), arXiv:1804.11326.
ZZFeatureMap
¶
ZZ feature map circuit builder.
Encodes classical data x into a quantum state using single-qubit Z rotations and two-qubit ZZ entangling gates, repeated reps times.
build_circuit(x, n_qubits, reps=2)
staticmethod
¶
Build the ZZ feature map circuit for data vector x.
Parameters¶
x : array of shape (n_qubits,)
Classical feature vector (values should be in [0, 2*pi]).
n_qubits : int
Number of qubits (must match len(x)).
reps : int
Number of repetitions of the feature map layer.
Returns¶
QuantumCircuit Qiskit circuit encoding x.
QuantumKernelClassifier
dataclass
¶
quantum_kernel(x1, x2, n_qubits, backend, reps=2)
¶
Compute the quantum kernel value between two data points.
Uses statevector overlap: k(x1, x2) = |<0|U^dag(x2) U(x1)|0>|^2.
Classifiers¶
classifiers
¶
Variational quantum classifiers for fraud detection.
Implements a parameterized variational quantum classifier (VQC) with angle encoding and a TwoLocal ansatz, trained via gradient-free optimization (COBYLA).
References¶
Schuld, Bocharov, Svore, Killoran, PRA 101, 032308 (2020).
VQCConfig
dataclass
¶
Configuration for a variational quantum classifier.
VariationalQuantumClassifier
¶
Variational quantum classifier with angle encoding and TwoLocal ansatz.
Features are encoded via R_Y rotations. The ansatz alternates layers of R_Y / R_Z single-qubit rotations with a ladder of CNOT gates. The model is trained by minimising cross-entropy loss using a gradient-free optimizer (COBYLA by default).
Quantum GAN¶
qgan
¶
qGANs for distribution loading (Zoufal et al., npj QI 5:103, 2019).
Implements a quantum Generative Adversarial Network where the generator is a parameterized quantum circuit and the discriminator is a classical neural network. The trained generator circuit can then be used to load arbitrary probability distributions into quantum states.
References¶
Zoufal, Lucchi, Woerner, npj Quantum Information 5:103 (2019), arXiv:1904.00043.
QGANConfig
dataclass
¶
Configuration for qGAN training.
QGANResult
dataclass
¶
Result from qGAN training.
QuantumGAN
¶
Quantum GAN for distribution loading.
The generator is a TwoLocal-style parameterized quantum circuit. The discriminator is a simple classical neural network implemented with numpy (no torch/tensorflow dependency).
train()
¶
Train the qGAN.
Uses finite-difference gradient estimation for the quantum generator and numpy-based SGD for the discriminator.
Reservoir Computing¶
reservoir
¶
Quantum reservoir computing for volatility forecasting (arXiv:2505.13933).
Implements a quantum reservoir based on a transverse-field Ising Hamiltonian. Input data is encoded via single-qubit rotations, the reservoir dynamics evolve the state, and expectation values are extracted as features for a classical ridge-regression readout layer.
References¶
Li et al., arXiv:2505.13933 (2025).
QuantumReservoirConfig
dataclass
¶
Configuration for quantum reservoir computing.
QuantumReservoir
¶
Quantum reservoir for time-series regression.
Uses a transverse-field Ising Hamiltonian as reservoir dynamics:
H = -J sum_{
Input data is encoded via R_Y rotations, then reservoir layers apply ZZ couplings and transverse-field X rotations. Expectation values of Pauli-Z operators serve as the feature vector for a linear readout.