hqnn-forge

Hybrid quantum-classical neural networks for tabular binary and multiclass classification: quantum encoding layers, hybrid classifiers, training with early stopping and imbalance-aware losses, noise models, quantum kernels and the diagnostics to judge them, built on PennyLane and PyTorch.

hqnn-forge is not on PyPI; install it from a clone of the repository:

git clone https://github.com/hqnn-forge/hqnn-forge.git
cd hqnn-forge
pip install -e ".[lightning]"    # the library, plus the fast C++ simulator

import torch
from hqnn_forge.models import HybridBinaryClassifier

model = HybridBinaryClassifier(n_input_features=30, n_qubits=8, n_layers=2)
probabilities = model.predict_proba(torch.randn(4, 30))

The README covers installation, device backends and the architectures in detail; this site is the API reference, generated from the docstrings.

API reference

Package	Contents
hqnn_forge.encoding	Quantum feature maps (angle, IQP, amplitude, data re-uploading) and the circuit pieces they share
hqnn_forge.models	The hybrid classifiers, their MLP control and a linear classifier
hqnn_forge.noise	Noise channels (depolarizing, amplitude and phase damping, bit and phase flip), post hoc and during training
hqnn_forge.kernels	Quantum fidelity kernels
hqnn_forge.diagnostics	Circuit summaries, gradient variance, Fisher spectrum, effective dimension, expressibility, entangling capability, separation measures of a fixed feature matrix
hqnn_forge.evaluation	Metrics, threshold search, statistical tests, plots
hqnn_forge.training	The training loop
hqnn_forge.data	Dataset loaders
hqnn_forge.preprocessing	PCA, cross-validation and fold-safe SMOTE
hqnn_forge.sklearn	The scikit-learn estimator
hqnn_forge.utils	Losses, checkpoints, ablation, evaluation-mode helper
hqnn_forge.circuits	Ansatz primitives
hqnn_forge.initializers	Small-angle weight initialisation
hqnn_forge.rydberg	Atom register, dense Rydberg Hamiltonian, dephasing (Lindblad) solver and fixed feature map (inputs to excitation probabilities) of a small neutral-atom array (model)
hqnn_forge.benchmark	Hybrid model against its matched classical control, on identical folds
hqnn_forge.experiment	JSON experiment records: save, load and rerun a benchmark