PredictiveCO Benchmark

Overview

PredictiveCO Benchmark is a benchmark testing framework for Predictive Combinatorial Optimization (Predictive‑CO). The framework aims to evaluate methods under two design principles: “Predict‑then‑Optimize (PtO)” and “Predict‑and‑Optimize (PnO)”.

Datasets

• Problem Types: Covers 8 problems, including a new industrial dataset for combinatorial advertising.
• Methods: Benchmark includes 11 existing PtO/PnO methods.

Modular Framework

• User Customization: Users can plug in their own problems, predictors, solvers, loss functions, and evaluation methods.

Usage

• Installation: Install locally with:

  pip install -e .

• Running Algorithms: Refer to the shell scripts in the “shells/benchmarks” directory.

Citation

bibtex @inproceedings{geng2024predictive, title={Benchmarking PtO and PnO Methods in the Predictive Combinatorial Optimization Regime}, author={Geng, Haoyu and Ruan, Hang and Wang, Runzhong and Li, Yang and Wang, Yang and Chen, Lei and Yan, Junchi}, booktitle={NeurIPS 2024 Datasets and Benchmarks Track}, year={2024} }

Terms of Use

By using this benchmark dataset, you agree to the terms of use specified in Appendix C.