This repo contains the code for the agents in the AAAI 2020 Oral paper "Theory-based Causal Transfer: Integrating Instance-level Induction and Abstract-level Structure Learning." For the paper and additional details, please see the project page.
It provides various agents to use with the OpenLock OpenAI Gym environment
You'll need to bring in the OpenLock repo into your PYTHONPATH for this repo. If you are using a virtualenv, you can add export PYTHONPATH="/path/to/OpenLock" to /bin/activate. Alternatively, you can python setup.py install OpenLock to add it as a package to your python environment.
- Clone this repo, cd into it, and run
python setup.py install - Follow OpenLock instructions here: https://github.com/mjedmonds/OpenLock
You can run any of the agents in the openlockagents directory. The main python file for each agent will end with _open_lock.py. To run the model presented in the paper, run:
python openlockagents/OpenLockLearner/main/openlock_learner_open_lock.py
If you use this repo, please cite our work:
@inproceedings{edmonds2020theory,
title={Theory-based Causal Transfer: Integrating Instance-level Induction and Abstract-level Structure Learning},
author={Edmonds, Mark and Ma, Xiaojian and Qi, Siyuan, and Zhu, Yixin and Lu, Hongjing and Zhu, Song-Chun},
booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
year={2020}
}