life1904/vgc-battle-logs
Dataset from the paper VGC-Bench: Towards Mastering Diverse Team Strategies in Competitive Pokémon. Code: vgc-bench Contains data for 18 Gen 9 Pokémon VGC formats, all filtered for Open Team Sheets (OTS). See: scrape_logs.py for the code that scraped these 906,870 logs from the Pokémon Showdown replay database 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… See the full description on the dataset page: https://huggingface.co/datasets/life1904/vgc-battle-logs.
Dataset from the paper VGC-Bench: Towards Mastering Diverse Team Strategies in Competitive Pokémon.
Code: vgc-bench
Contains data for 18 Gen 9 Pokémon VGC formats, all filtered for Open Team Sheets (OTS). See:
- scrape_logs.py for the code that scraped these _906,870_ logs from the Pokémon Showdown replay database
- 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 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>"),
...
}Cite us
@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}
}