Example Game 1 |
---|
Instead of generating a new level on every reset operation, a pregenerated level is choosen randomly. When Boxoban is run for the first time the levels are downloaded from DeepMinds Github repository and stored in the folder .sokoban_cache.
In case you use the Boxoban levels for your research, the authors of the boxoban-levels repository ask you to cite their work as follows:
@misc{boxobanlevels,
author = {Arthur Guez, Mehdi Mirza, Karol Gregor, Rishabh Kabra, Sebastien Racaniere, Theophane Weber, David Raposo, Adam Santoro, Laurent Orseau, Tom Eccles, Greg Wayne, David Silver, Timothy Lillicrap, Victor Valdes},
title = {An investigation of Model-free planning: boxoban levels},
howpublished= {https://github.com/deepmind/boxoban-levels/},
year = "2018",
}
Dedending on your use case, please consider the licence of DeepMinds repository.
Same rules as the regular Sokoban game.
Through the API we provide access to 'unfiltered' and 'medium' levels. The unflitered levels are split into a train, test, and validation set. Whereas the medium levels are only split into train and validation. For more details see DeepMind's README.md. Of course all configurations can be rendered as TinyWorld.
Room Id | Grid-Size | Grid-Size | Pixels | #Boxes |
---|---|---|---|---|
Boxoban-Train-v0 | unfiltered | 10x10 | 112x112 | 4 |
Boxoban-Test-v0 | unfiltered | 10x10 | 112x112 | 4 |
Boxoban-Val-v0 | unfiltered | 10x10 | 112x112 | 4 |
Boxoban-Train-v1 | medium | 10x10 | 112x112 | 4 |
Boxoban-Val-v1 | medium | 10x10 | 112x112 | 4 |