lingweizhu/disc

Dimension-Wise Importance Sampling Weight Clipping

This repository is an implementation of Dimension-Wise Importance Sampling Weight Clipping for Sample-Efficient Reinforcement Learning (ICML 2019)

@article{han2019dimension,
  title={Dimension-Wise Importance Sampling Weight Clipping for Sample-Efficient Reinforcement Learning},
  author={Han, Seungyul and Sung, Youngchul},
  journal={arXiv preprint arXiv:1905.02363},
  year={2019}
}

Dependencies

The implementation is based on Open AI baselines (https://github.com/openai/baselines)

It requires Python 3.*/Tensorflow.

Local Installation

1.Install Anaconda & Mujoco 131

2.Unzip disc.zip into your installation path

'''
cd <installation_path>
unzip disc.zip
cd disc
'''

3.Create environment

'''
conda create -n disc python=3.5
source activate disc
python setup.py install
'''

Training on Mujoco tasks

'''
cd <installation_path>/disc
source activate disc
python -m baselines.run_disc --env=Humanoid-v1 --num_timesteps 1e7 --log_dir ./Results/Humanoid-v1
'''