tig-foundation/tig-challenges-data

---
# Dataset card metadata — https://huggingface.co/docs/hub/datasets-cards
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)](https://tig.foundation). 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](https://github.com/tig-foundation/tig-challenges)**. This repo is only storage for data and published run outputs.

## What is here

| Area | Role |
|------|------|
| **Instances** | `<challenge>/test/` — `*.txt` problem files under `academic/` (literature sets) and `tig/` (TIG tracks). |
| **Published best-known values (academic)** | CSV files next to those instances, e.g. `knapsack/test/academic/standard-QKP/standard_qkp_bks.csv`, `job_scheduling/test/academic/bks.csv`, `vehicle_routing/test/academic/bks.csv`, and the other knapsack academic BKS CSVs alongside their splits. |
| **Long-run summary** | `<challenge>/results/summary_long_run_v*.csv` — rolled-up metrics from the reference run (where that challenge has been published). |
| **Convergence plots** | `<challenge>/results/plots/*.png` — figures from that run (e.g. average gap vs. time), when present. |
| **Solutions** | `<challenge>/results/solutions/` — final `.solution` files and optional time-stamped snapshots (`.solution.<seconds>`), laid out to mirror the corresponding instance paths under `test/`, plus `runs.csv` where exported. |

Challenges: **knapsack**, **job_scheduling**, **vehicle_routing**. What appears under each challenge’s `results/` folder depends on what has been published for that challenge (summaries, plots, solutions).

## Hugging Face

Dataset hub entry: [tig-foundation/tig-challenges-data](https://huggingface.co/datasets/tig-foundation/tig-challenges-data). You can mirror this tree with:

```bash
git clone https://huggingface.co/datasets/tig-foundation/tig-challenges-data
```

## Citation

```bibtex
@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}
}
```