https://github.com/leo27945875/adversarial-soft-advantage-fitting-1

https://github.com/leo27945875/adversarial-soft-advantage-fitting-1

deep-learning generative-adversarial-network imitation-learning reinforcement-learning

Last synced: 7 days ago
JSON representation

• Host: GitHub
• URL: https://github.com/leo27945875/adversarial-soft-advantage-fitting-1
• Owner: leo27945875
• Created: 2021-09-09T16:53:49.000Z (over 3 years ago)
• Default Branch: master
• Last Pushed: 2021-11-17T17:25:25.000Z (about 3 years ago)
• Last Synced: 2024-12-01T15:36:05.569Z (2 months ago)
• Topics: deep-learning, generative-adversarial-network, imitation-learning, reinforcement-learning
• Language: Jupyter Notebook
• Homepage:
• Size: 75.2 MB
• Stars: 1
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • Changelog: history/ASAF1_128_Ant-v2.pkl

Awesome Lists containing this project

README

        
# Adversarial Soft Advantage Fitting - 1 (ASAF-1)

Imitation learning without policy optimization !

## Introduction (In Traditional Chinese):

https://www.notion.so/Adversarial-Soft-Advantage-Fitting-441698eb0ccb40eab4f59275d637466a

## Description

### Preparing expert demos:

Every expert demo (state-action pairs) file must be a pickle file and in this form: [[np.array([state0]), np.array([action0])], [np.array([state1]), np.array([action1])], ...]

### Training:

Adjust the parameters in ./src/train.py and then run it.

### Testing:

Adjust the parameters in ./src/test.py and then run it.

## Experiment

![](./image/SS%201.png)  

## Reference

https://arxiv.org/abs/2006.13258