Introduction

Welcome to RL4Ising!

RL4Ising, or Reinforcement Learning for Ising Model, is an open-source dataset and benchmark suite for Ising models. We aim to curate a public dataset of Ising models, provide a comprehensive benchmark of state-of-the-art (SOTA) reinforcement learning (RL) algorithms alongside an industry-standard solver baseline, and offer detailed tutorials for key RL algorithms.

Through the open-source dataset and benchmark, RL4Ising offers a diverse set of Ising model instances, ease of reproducibility, and strong performace evaulation:

• Diversity: RL4Ising provides over 170,000 Ising instances spanning different lattice types, 1 to 4 dimensions, spin interactions.

• Reproducibility: RL4Ising provides tutorials for the solvers and methods used to obtain benchmarked data.

• Performace Evaulation: RL4Ising provides cross-evaluation between SOTA algorithms and RL solvers.

[Lin, L., Wang, Z., Mac Entee, H., Zhao, X., & Liu, X.-Y. Reinforcement Learning for Ising Models: Datasets and Benchmark, NeurIPS ML4PS Workshop, 2025.](https://ml4physicalsciences.github.io/2025/files/NeurIPS_ML4PS_2025_245.pdf)