qml-pennylane 0.1.8

Modular PennyLane-based quantum machine learning suite for classification, regression, and quantum kernel methods.

Quantum Machine Learning

Modular PennyLane-based quantum machine learning library implementing reusable workflows for:

• Variational quantum classification (VQC)
• Variational quantum regression (VQR)
• Quantum kernel methods
• Trainable quantum kernels (kernel-target alignment)
• Quantum metric learning (trainable embedding geometry)
• Classical baseline models
• Deterministic benchmark utilities

The repository follows a package-first design:

• algorithms implemented in qml/
• notebooks act as thin clients
• experiments produce reproducible outputs
• consistent plotting and result structures
• deterministic execution via explicit seeds

---

Installation

Clone and install in editable mode:

pip install -e .

Install development tools:

pip install -e ".[dev]"

Requirements:

• Python ≥ 3.10 • PennyLane ≥ 0.34 • NumPy ≥ 1.24 • scikit-learn ≥ 1.3 • matplotlib ≥ 3.7

---

Quick start

Variational quantum classifier

from qml.classifiers import run_vqc

result = run_vqc(
    n_samples=200,
    n_layers=2,
    steps=50,
    plot=True,
)

---

Variational quantum regression

from qml.regression import run_vqr

result = run_vqr(
    n_samples=200,
    n_layers=2,
    steps=50,
    plot=True,
)

---

Quantum kernel classifier

from qml.kernel_methods import run_quantum_kernel_classifier

result = run_quantum_kernel_classifier(
    n_samples=200,
    plot=True,
)

---

Trainable quantum kernel (kernel-target alignment)

from qml.trainable_kernels import run_trainable_quantum_kernel_classifier

result = run_trainable_quantum_kernel_classifier(
    n_samples=200,
    steps=50,
    plot=True,
)

---

Quantum metric learning

from qml.metric_learning import run_quantum_metric_learner

result = run_quantum_metric_learner(
    samples=200,
    layers=2,
    steps=50,
    plot=True,
)

Learns a trainable embedding circuit using contrastive supervision:

• same-class samples mapped closer together
• different-class samples separated in feature space

Classification is performed via nearest-centroid prediction in the learned embedding.

---

All workflows return structured dictionaries containing:

• training metrics • predictions • learned parameters • configuration metadata

---

Noise-aware execution (finite shots)

Quantum circuits can be evaluated either analytically or with finite sampling.

Finite-shot execution uses:

qml.set_shots(qnode, shots)

Example:

result = run_vqc(
    n_samples=200,
    n_layers=2,
    steps=50,
    shots=128,
)

Trainable kernel workflows support separate shot settings:

result = run_trainable_quantum_kernel_classifier(
    n_samples=200,
    shots_train=64,
    shots_kernel=256,
)

All workflows remain deterministic when a fixed seed is provided.

---

Benchmark framework

Benchmark utilities compare quantum and classical models across multiple seeds.

Example:

from qml.benchmarks import compare_classification_models

result = compare_classification_models(
    models=[
        "vqc",
        "quantum_kernel",
        "trainable_quantum_kernel",
        "logistic_regression",
        "svm_classifier",
    ],
    seeds=[123, 456],
)

---

Model-specific configuration

Benchmarks accept per-model kwargs:

result = compare_classification_models(
    models=[
        "vqc",
        "quantum_kernel",
        "trainable_quantum_kernel",
    ],
    seeds=[123],
    model_kwargs={
        "vqc": {"shots": 128},

        "quantum_kernel": {"shots": 256},

        "trainable_quantum_kernel": {
            "shots_train": 64,
            "shots_kernel": 256,
        },
    },
)

Result structure remains consistent across models.

---

Classical baselines

Included reference models:

• logistic regression • ridge regression • support vector machine • multilayer perceptron

These provide performance context for quantum models.

---

Command line interface

Run workflows directly:

python -m qml vqc --steps 50 --plot
python -m qml regression --steps 50 --plot
python -m qml kernel --plot
python -m qml trainable-kernel --steps 50 --plot
python -m qml metric-learning --steps 50 --plot

Run benchmarks:

python -m qml benchmark classification \
    --models vqc quantum_kernel svm_classifier logistic_regression \
    --seeds 123 456

python -m qml benchmark regression \
    --models vqr ridge_regression mlp_regressor \
    --seeds 123 456

CLI outputs include:

• training metrics • test metrics • final loss • saved plots (optional)

---

Documentation

Core documentation:

• THEORY.md — mathematical background • USAGE.md — API examples

Algorithm notes:

• docs/qml/variational_quantum_classifier.md • docs/qml/variational_regression.md • docs/qml/quantum_kernels.md

Example notebooks:

• quantum_variational_classifier.ipynb • quantum_regressor.ipynb • quantum_kernel_classifier.ipynb • classical_vs_quantum_classifier.ipynb

---

Repository structure

qml/

    ansatz.py
        parameterised circuit templates

    embeddings.py
        feature encoding circuits

    classifiers.py
        variational quantum classification workflows

    regression.py
        variational quantum regression workflows

    kernel_methods.py
        quantum kernel workflows

    trainable_kernels.py
        kernel-target alignment optimisation

    metric_learning.py
        contrastive quantum embedding optimisation

    classical_baselines.py
        logistic, ridge, svm, mlp

    benchmarks.py
        multi-seed benchmark utilities

    training.py
        hybrid optimisation loops

    metrics.py
        evaluation metrics

    losses.py
        objective functions

    data.py
        dataset generation utilities

    visualize.py
        plotting utilities

    io_utils.py
        reproducible saving utilities


notebooks/

    examples implemented as thin package clients


tests/

    smoke tests
    deterministic benchmarks


docs/

    theory notes and algorithm descriptions


results/

    saved experiment outputs (gitignored)


images/

    generated plots (gitignored)

---

Design principles

Package-first architecture

Core implementations live in:

qml.*

Notebooks import public APIs rather than defining circuits inline.

---

Deterministic workflows

Reproducibility is prioritised:

• explicit random seeds • deterministic dataset generation • reproducible optimisation • consistent JSON outputs • deterministic finite-shot execution

---

Minimal abstractions

Shared infrastructure intentionally remains lightweight:

• small set of embeddings • hardware-efficient ansatz • simple optimisation loops • consistent plotting utilities

---

Current algorithms

Variational quantum classifier

Binary classification using:

• angle embedding • hardware-efficient ansatz • cross-entropy loss

---

Variational quantum regression

Continuous prediction using:

• angle embedding • expectation-value outputs • mean squared error

---

Quantum kernel classifier

Support vector machine using quantum feature maps:

$$ K(x_i, x_j)

|\langle \phi(x_i) | \phi(x_j) \rangle|^2 $$

---

Trainable quantum kernel

Kernel alignment objective:

$$ \max_\theta ; \frac{ \langle K_\theta, Y \rangle_F }{ |K_\theta|_F |Y|_F } $$

where:

• $K_\theta$ is the quantum kernel matrix • $Y$ is the label similarity matrix

---

Quantum metric learning

Supervised embedding optimisation using contrastive loss:

$$ L = y d^2 + (1 - y)\max(0, m - d)^2 $$

where:

• $d$ is distance between learned embeddings
• $y \in {0,1}$ indicates whether samples share a class
• $m$ is a separation margin

The learned embedding is used for classification via nearest-centroid prediction in feature space.

Supports:

• trainable data re-uploading embeddings
• stochastic pair sampling
• deterministic optimisation via fixed seeds
• consistent evaluation pipeline with other models

---

Development workflow

Run tests:

pytest

Format code:

black .
ruff check .

Run module:

python -m qml

---

Author

Sid Richards

LinkedIn: https://www.linkedin.com/in/sid-richards-21374b30b/

GitHub: https://github.com/SidRichardsQuantum

---

License

MIT License — see LICENSE