https://github.com/wongongv/PolicyGradient_in_tensorflow2.0

Implemented Policy Gradient in Tensorflow2.0
https://github.com/wongongv/PolicyGradient_in_tensorflow2.0

Last synced: 10 days ago
JSON representation

Implemented Policy Gradient in Tensorflow2.0

• Host: GitHub
• URL: https://github.com/wongongv/PolicyGradient_in_tensorflow2.0
• Owner: wsonv
• License: mit
• Created: 2019-04-29T05:17:29.000Z (almost 6 years ago)
• Default Branch: master
• Last Pushed: 2019-06-24T08:22:11.000Z (over 5 years ago)
• Last Synced: 2024-08-10T11:01:21.495Z (6 months ago)
• Language: Python
• Homepage:
• Size: 167 KB
• Stars: 3
• Watchers: 0
• Forks: 1
• Open Issues: 1
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

• awesome-tensorflow-2 - Implemented Policy Gradient in Tensorflow2.0

README

        
# PolicyGradient_in_tensorflow2.0

Implementation of Policy Gradient in Tensorflow2.0

* This code referenced skeleton code, which is in TensorFlow 1.x, offered in the course [CS294-112 github](https://github.com/berkeleydeeprlcourse/homework/tree/master/hw2)

* Implemented for both continuous and discrete action spaces with reward-to-go option.

* Implemented q function normalization to decrease variation.

## Prerequisites

* python3

* Tensorflow2.0

* Tensorflow_probability (you need to use nightly build to use it in TF2.0)

* numpy

* gym 

## Example Usage

### Training

*available arguments : --discount, --n_experiment, --n_iter, --seed, --batch, --learning_rate, --exp_name. --render

	$ python PG_in_tf2.0.py CartPole-v0 --exp_name Test_disc --seed 1 --render

	$ python PG_in_tf2.0.py Pendulum-v0 --exp_name Test_cont --seed 1

### Plotting

	$ python plot.py Data/your_exp_name

	$ python plot.py Data/your_exp_name -f

## Example Result

CartPole-v0

iteration : 100

number of experiments : 2

![CartPole](result/CartPole-full.png)

## Author

Wonjun Son / [Github](https://github.com/wongongv)