---
license: mit
task_categories:
- reinforcement-learning
tags:
- GWM
- Pacman
size_categories:
- 100K<n<1M
language:
- en
---

This dataset contains state-action pairs of three different Pacman levels, each located in its own tar.gz.

Each `tar.gz` contains `output/`, `dataset.csv`, and `results.txt`.
- `output/` contains all the images/frames generated by an RL agent based on [Taylor Downey's RL Agent](https://github.com/tadowney/ai_pacman).
- `dataset.csv` contains state-action pairs that match each frame to an action of either North, South, East, West, or Stop. There is also a column indicating whether a new game has started.
- `results.txt` contains the results of either `Win` or `Lose` for every game.