https://github.com/openai/imitation

Code for the paper "Generative Adversarial Imitation Learning"
https://github.com/openai/imitation

paper

Last synced: 27 days ago
JSON representation

Code for the paper "Generative Adversarial Imitation Learning"

• Host: GitHub
• URL: https://github.com/openai/imitation
• Owner: openai
• License: mit
• Created: 2016-06-10T20:40:03.000Z (about 8 years ago)
• Default Branch: master
• Last Pushed: 2018-11-22T00:39:47.000Z (over 5 years ago)
• Last Synced: 2024-02-21T23:37:56.828Z (4 months ago)
• Topics: paper
• Language: Python
• Homepage: https://arxiv.org/abs/1606.03476
• Size: 29.7 MB
• Stars: 661
• Watchers: 178
• Forks: 194
• Open Issues: 9
• Metadata Files:
  • Readme: README.rst
  • License: LICENSE

Lists

• awesome-GAN - imitation - A implementation of the paper [Generative Adversarial Imitation Learning](https://arxiv.org/abs/1606.03476) (Codes)
• awesome-stars - imitation - Code for the paper "Generative Adversarial Imitation Learning" (Python)

README

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

=========================================

Generative Adversarial Imitation Learning

=========================================

-----------------------------------------

Jonathan Ho and Stefano Ermon

-----------------------------------------

Contains an implementation of Trust Region Policy Optimization (Schulman et al., 2015).

Dependencies:

* OpenAI Gym >= 0.1.0, mujoco_py >= 0.4.0

* numpy >= 1.10.4, scipy >= 0.17.0, theano >= 0.8.2

* h5py, pytables, pandas, matplotlib

Provided files:

* ``expert_policies/*`` are the expert policies, trained by TRPO (``scripts/run_rl_mj.py``) on the true costs

* ``scripts/im_pipeline.py`` is the main training and evaluation pipeline. This script is responsible for sampling data from experts to generate training data, running the training code (``scripts/imitate_mj.py``), and evaluating the resulting policies.

* ``pipelines/*`` are the experiment specifications provided to ``scripts/im_pipeline.py``

* ``results/*`` contain evaluation data for the learned policies