Circuit Design based on Feature Similarity for Quantum Generative Modeling

This repository includes the source code for simulations carried out in Ref. [1].

First setup

Install dependencies or open in devcontainer to automatically set up the envorinment.

pip install -r requirements.txt

Download JGB dataset from: https://www.mof.go.jp/english/policy/jgbs/reference/interest_rate/index.htm and insert in this project data/jgbcme_all.csv

Project Structure

• .devcontainer/ for running the devcontainer
• data/ need to create this folder by yourself when adding data files
• results/
  • runs/... results of the numerical simulations
  • plot_figures.ipynb plot the figures for the paper
• src/ code for running simulations
  • src/conf/config.yaml configuration for the simulation
  • cost.py loss functions, gradients, and meta-optimizer
  • data.py dataclasses and dataloader
  • decomposition.py decompose MPS to PQC by Yuki Sato @yksat
  • extension.py functions for extending the decomposed circuit
  • mps.py train MPS by @congzlwag
  • qcbm.py train QCBM using qiskit
  • setup.py setup the whole simulation from config including pretraining
  • utils.py general utility functions and plotting

Singlerun

Running a single simulation based on current config file src/conf/config.yaml

python -m src

Results will be saved in outputs/

Multirun

Running multiple simulations sequentially based on same config file but changing config parameters e.g. the extension method

python -m src --multirun extension=none,metric_based,all_to_all

Important for running simulations with random extension is changing the random_seed, which is somehow fixed when using multirun from hydra

python -m src --multirun random_seed=42,43,44,45,46

Results will be saved in multirun/

Testing and code coverage

pytest src/tests
pytest --cov=src

Reference

[1] Mathis Makarski, Jumpei Kato, Yuki Sato, Naoki Yamamoto, Circuit Design based on Feature Similarity for Quantum Generative Modeling, arXiv preprint (2025)