life1904/vgc-battle-logs

---
license: mit
task_categories:
- reinforcement-learning
size_categories:
- 100K<n<1M
---

Dataset from the paper [VGC-Bench: Towards Mastering Diverse Team Strategies in Competitive Pokémon](https://arxiv.org/abs/2506.10326).

Code: [vgc-bench](https://github.com/cameronangliss/vgc-bench)

Contains data for 18 Gen 9 Pokémon VGC formats, all filtered for Open Team Sheets (OTS). See:
- [scrape_logs.py](https://github.com/cameronangliss/vgc-bench/blob/main/vgc_bench/scrape_logs.py) for the code that scraped these __906,870__ logs from the [Pokémon Showdown replay database](https://replay.pokemonshowdown.com)
- [logs2trajs.py](https://github.com/cameronangliss/vgc-bench/blob/main/vgc_bench/logs2trajs.py) for the code that reads the logs from player 1 and 2's perspective, yielding __1,813,688__ trajectories (__99.997%__ successful log read rate), or __16,832,842__ transitions
- [pretrain.py](https://github.com/cameronangliss/vgc-bench/blob/main/vgc_bench/pretrain.py) for the code that uses those ~17M transitions to train a policy with behavior cloning

Each provided json file contains data in the following format:
```
{
"<battle-id>": ("<time-of-battle-in-epoch-seconds>", "<battle-log>"),
...
}
```

| VGC Format | Latest Log (GMT) | Total Logs | Unrated | 1000-1099 | 1100-1199 | 1200-1299 | 1300-1399 | 1400-1499 | 1500-1599 | 1600-1699 | 1700-1799 | 1800+ |
|------------------|----------------------|------------|---------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|-------|
| Reg D (2023) | 03-19-2026, 08:55:38 | 579 | 1,158 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| Reg F (2024) | 05-28-2024, 23:16:45 | 16,445 | 32,890 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| Reg F BO3 (2024) | 05-27-2024, 16:29:21 | 66 | 132 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| Reg G (2024) | 04-03-2026, 17:00:49 | 22,418 | 44,836 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| Reg G BO3 (2024) | 09-01-2024, 00:58:10 | 58 | 116 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| Reg H (2024) | 01-01-2025, 17:44:04 | 16,020 | 31,520 | 44 | 117 | 88 | 97 | 89 | 59 | 16 | 9 | 1 |
| Reg H BO3 (2024) | 12-15-2024, 20:02:50 | 43 | 86 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| Reg G (2025) | 05-01-2025, 16:16:30 | 16,451 | 19,350 | 1,889 | 3,842 | 2,832 | 2,691 | 1,478 | 589 | 193 | 36 | 2 |
| Reg G BO3 (2025) | 05-01-2025, 16:43:52 | 191,016 | 41,484 | 78,396 | 111,466 | 72,678 | 48,962 | 21,916 | 6,223 | 887 | 20 | 0 |
| Reg H (2025) | 12-02-2025, 04:11:56 | 7,865 | 8,936 | 1,135 | 1,832 | 1,259 | 1,277 | 825 | 386 | 74 | 6 | 0 |
| Reg H BO3 (2025) | 12-02-2025, 06:13:04 | 154,800 | 31,774 | 76,088 | 94,643 | 56,144 | 33,921 | 13,209 | 3,382 | 392 | 47 | 0 |
| Reg I (2025) | 01-05-2026, 18:41:24 | 16,268 | 19,786 | 1,988 | 3,455 | 2,624 | 2,360 | 1,458 | 669 | 179 | 15 | 2 |
| Reg I BO3 (2025) | 09-01-2025, 15:21:06 | 183,267 | 43,314 | 81,082 | 110,912 | 68,438 | 42,764 | 15,506 | 3,789 | 633 | 96 | 0 |
| Reg J BO3 (2025) | 01-06-2026, 02:46:38 | 8,051 | 16,102 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| Reg F (2026) | 04-03-2026, 18:52:11 | 14,994 | 19,342 | 1,934 | 2,746 | 1,858 | 1,803 | 1,312 | 642 | 290 | 57 | 4 |
| Reg F BO3 (2026) | 04-02-2026, 06:57:34 | 224,879 | 54,858 | 91,238 | 125,714 | 87,796 | 60,184 | 22,637 | 6,469 | 820 | 42 | 0 |
| Reg I (2026) | 04-03-2026, 18:56:20 | 2,336 | 2,876 | 348 | 648 | 430 | 253 | 93 | 21 | 3 | 0 | 0 |
| Reg I BO3 (2026) | 04-03-2026, 19:35:46 | 31,314 | 5,038 | 22,363 | 22,545 | 8,976 | 3,019 | 610 | 67 | 10 | 0 | 0 |

# Cite us

```bibtex
@inproceedings{anglissvgc,
title={VGC-Bench: Towards Mastering Diverse Team Strategies in Competitive Pok{\'e}mon},
author={Angliss, Cameron L and Cui, Jiaxun and Hu, Jiaheng and Rahman, Arrasy and Stone, Peter},
booktitle={The 25th International Conference on Autonomous Agents and Multi-Agent Systems}
}
```