Code for reproducing Forward Prediction for Physical Reasoning.
A Conda virtual enviroment is provided contianing all necessary dependencies.
git clone https://github.com/facebookresearch/phyre-fwd.git
cd phyre-fwd
conda env create -f env.yml && conda activate phyre-fwd
pushd src/python && pip install -e . && popd
pip install -r requirements.agents.txt
Make a directory somewhere with enough space for trained models, and symlink it to agents/outputs
.
For training the methods, and available pre-trained models, see agents.
PHYRE forward agents are released under the Apache license. See LICENSE for additional details.
If you use phyre-fwd
or the baseline results, please cite it:
@article{girdhar2020forward,
title={Forward Prediction for Physical Reasoning},
author={Rohit Girdhar and Laura Gustafson and Aaron Adcock and Laurens van der Maaten},
year={2020},
journal={arXiv:2006.10734}
}