Datasets:
pretty_name: TIG Challenges Data
license: mit
language:
- en
tags:
- optimization
- benchmark
- combinatorial-optimization
- tig
size_categories:
- 10K<n<100K
TIG Challenges Data
This repository holds benchmark instances, reference objective data for academic splits, and artifacts from long reference runs (aggregated summaries, convergence plots, and solution files) for the combinatorial challenges used in The Innovation Game (TIG). It is meant as a stable place to download benchmarks and compare against those reference outputs.
Solver code, formats, and how to run or evaluate algorithms live in tig-foundation/tig-challenges. This repo is only storage for data and published run outputs.
What is here
Layout matches this checkout (and may differ from older Hub revisions that used a test/ prefix):
| Area | Role |
|---|---|
| Instances | <challenge>/instances/ — *.txt instances under splits such as TIG/, standard-QKP/, TeamFormation-QKP/, Dispersion-QKP/ (knapsack), tig/, fjsp/, jsp/ (job scheduling), TIG/ (vehicle routing), etc. |
| Best-known values | knapsack/instances/bks.csv, job_scheduling/instances/bks.csv, vehicle_routing/instances/bks.csv, plus any BKS CSVs alongside specific academic splits under instances/. |
| Plots | <challenge>/plots/*.png — convergence or summary figures when present. |
| Solutions | <challenge>/solutions/ — published .solution files (and optional snapshots) mirroring instance paths where bundled, plus any exported runs.csv. |
Challenges: knapsack, job_scheduling, vehicle_routing. Not every challenge includes all of plots, solutions, and every split in a given clone.
Hugging Face
Dataset hub entry: tig-foundation/tig-challenges-data. You can mirror this tree with:
git clone https://huggingface.co/datasets/tig-foundation/tig-challenges-data
Citation
@misc{tig-challenges-2026,
title={TIG Challenges: Benchmarks for AI-Driven Algorithm Discovery},
author={{TIG Foundation}},
year={2026},
url={https://huggingface.co/datasets/tig-foundation/tig-challenges-data}
}