Datasets:
license: mit
task_categories:
- reinforcement-learning
tags:
- imitation_learning
- gymnasium
- agents
pretty_name: Labyrinth 5x5 Dataset
size_categories:
- 1K<n<10K
language:
- en
Labyrinth 5x5 Dataset
This dataset is for the gymnasium-base environment Labyrinth-v0.
It contains obs, actions, rewards, episodeed_starts and info for each entry based on 32 train, 32 validation and 32 test different structures.
Dataset Details
Dataset Description
The dataset consists of the shortest route for each Labyrinth structure over 32 different structures for 3 different splits (train, validation and test).
Each entry cosists of:
obs (str): path to the image of the observation.
action (int): the action for that observation (0-4).
reward (float): the reward for that observation.
episode_starts (bool): whether that observation was the initial timestep for that path.
info (str): the custom setting language to load that structure into the environment if needed.
The dataset has 4 different splits (train, validation, test, and random).
The train, validation and test are 100 different structures for training, validating and testing, respectively.
The random split is for imitation learning and offline inverse reinforcement learning methods that require random samples to start (i.e., BCO, IUPE, SAIL).
Usage
Feel free to download all files from this dataset and use it as you please.
If you are interested in using our PyTorch Dataset implementation, feel free to check the IL Datasets project.
There, we implement a base Dataset that downloads this dataset and all other datasets directly from HuggingFace.
The BaselineDataset also allows for more control over the splits and how many episodes you want to use (in cases where the 100 episodes aren't necessary).
Citation