Status: Archive (code is provided as-is, no updates expected)

Evolved Policy Gradients (EPG)

The paper is located at https://arxiv.org/abs/1802.04821. A demonstration video can be found at https://youtu.be/-Z-ieH6w0LA.

Houthooft, R., Chen, R. Y., Isola, P., Stadie, B. C., Wolski, F., Ho, J., Abbeel, P. (2018). Evolved Policy Gradients. arXiv preprint arXiv:1802.04821.

Installation

Install Anaconda:

curl -o /tmp/miniconda.sh https://repo.continuum.io/miniconda/Miniconda3-latest-MacOSX-x86_64.sh
bash /tmp/miniconda.sh
conda create -n epg python=3.6.1
source activate epg

Install necessary OSX packages for MPI:

brew install open-mpi

Install necessary Python packages:

pip install mpi4py==3.0.0 scipy \
pandas tqdm joblib cloudpickle == 0.5.2 \
progressbar2 opencv-python flask >= 0.11.1 matplotlib pytest cython \
chainer pathos mujoco_py 'gym[all]'

Running

First go to the EPG code folder:

cd <path_to_EPG_folder>

Then launch the entry script:

PYTHONPATH=. python epg/launch_local.py

Experiment data is saved in <home_dir>/EPG_experiments/<month>-<day>/<experiment_name>.

Testing

First, set theta_load_path = '<path_to_theta.npy>/theta.npy' in launch_local.py according to the theta.npy obtained after running the launch_local.py script. This file should be located in /<home_dir>/EPG_experiments/<month>-<day>/<experiment_name>/thetas/.

Then run:

PYTHONPATH=. python epg/launch_local.py --test true

Visualizing experiment data

Assuming the experiment data is saved in <home_dir>/EPG_experiments/<month>-<day>/<experiment_name>, run:

PYTHONPATH=. python epg/viskit/frontend.py <home_dir>/EPG_experiments/<month>-<day>/<experiment_name>

Then go to http://0.0.0.0:5000 in your browser.

Viskit sourced from

Duan, Y., Chen, X., Houthooft, R., Schulman, J., Abbeel, P. "Benchmarking Deep Reinforcement Learning for Continuous Control". Proceedings of the 33rd International Conference on Machine Learning (ICML), 2016.

BibTeX entry

@article{Houthooft18Evolved,
author = {Houthooft, Rein and Chen, Richard Y. and Isola, Phillip and Stadie, Bradly C. and Wolski, Filip and Ho, Jonathan and Abbeel, Pieter},
title = {Evolved Policy Gradients},
journal={arXiv preprint arXiv:1802.04821},
year = {2018}}