Quantum Graph Connectivity Classifier

This project compares classical and quantum machine learning models for learning the connectivity property of graphs. It evaluates expressivity and performance of:

• Classical multivariate polynomial classifiers
• Quantum data re-uploading models using different variational ansätze

The script will:

• Load or generate graph data
• Train both classical and quantum models (depending on config)
• Plot training and test loss/accuracy with an annotated configuration box in the corner
• Save performance plots to the images/ directory

📁 Project Structure

├── data/
│ └── data_gen.py # Graph data generation code
├── models/
│ ├── classical_model.py # Classical polynomial model
│ └── quantum_model.py # Quantum model with Pennylane
├── images/ # Results generated across different runs
├── utils.py # Config parser and helper functions
├── plots.py # Plot results and visualise circuit
├── config.json # Configuration file (grid of settings)
├── main.py
└── README.md

⚙️ Configuration

All experiments are controlled through the config.json file. Multiple values in a list mean different iterations of execution will be done using each config. The following parameters can be set:

Option	Default	Description
generate_data	[false]	Controls whether the graph data is regenerated or reused from a previous run.
n_graphs	[200, 600]	Number of graphs generated using the Erdős–Rényi model.
n_nodes	[4, 5, 6, 7, 8, 9]	Number of nodes generated in the graph.
batch_size	[16]	Batch size for training the model.
learning_rate	[0.1]	Learning rate for the model.
n_layers	[1, 2, 3, 4, 5]	For Quantum, this is the number of layers of data encoding and variational circuit. For Classical, this is the degree of the multivariate polynomial.
epochs	[20, 50]	Number of epochs for the model to train.
ml_model	["classical", "quantum"]	Chooses which model to train on.
variational_ansatz	["rx", "rx_ry", "rx_ry_rz", "rx_ry_rz_ising"]	(Quantum only) Chooses which ansatz to train on, including the proposed rx_ry_rz_ising ansatz.
use_encoding_param	[false, true]	(Quantum only) Chooses if the graph uploading part of the circuit is parametrized or not.