https://github.com/frgfm/drlnd-p3-collaboration-competition
Collaboration and competition project of Udacity Deep Reinforcement Learning Nanodegree
https://github.com/frgfm/drlnd-p3-collaboration-competition
actor-critic ddpg-algorithm deep-reinforcement-learning experience-replay pytorch unity
Last synced: 2 months ago
JSON representation
Collaboration and competition project of Udacity Deep Reinforcement Learning Nanodegree
- Host: GitHub
- URL: https://github.com/frgfm/drlnd-p3-collaboration-competition
- Owner: frgfm
- License: mit
- Created: 2019-11-12T23:49:15.000Z (over 5 years ago)
- Default Branch: master
- Last Pushed: 2019-11-13T16:53:43.000Z (over 5 years ago)
- Last Synced: 2024-12-29T17:41:32.223Z (4 months ago)
- Topics: actor-critic, ddpg-algorithm, deep-reinforcement-learning, experience-replay, pytorch, unity
- Language: Python
- Size: 214 KB
- Stars: 0
- Watchers: 2
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Tennis coach
[](LICENSE) [](https://www.codacy.com/manual/fg/drlnd-p3-collaboration-competition?utm_source=github.com&utm_medium=referral&utm_content=frgfm/drlnd-p3-collaboration-competition&utm_campaign=Badge_Grade) [](https://circleci.com/gh/frgfm/drlnd-p3-collaboration-competition)
This repository is an implementation of DDPG agents for the collaboration and competition project of Udacity Deep Reinforcement Learning nanodegree, in the tennis game provided by unity.
## Table of Contents
- [Environment](#environment)
- [Getting Started](#getting-started)
- [Prerequisites](#prerequisites)
- [Installation](#installation)
- [Usage](#usage)
- [Credits](#credits)
- [License](#license)
## Environment
In this environment, two agents control rackets to bounce a ball over a net. If an agent hits the ball over the net, it receives a reward of +0.1. If an agent lets a ball hit the ground or hits the ball out of bounds, it receives a reward of -0.01. Thus, the goal of each agent is to keep the ball in play.
The observation space consists of 8 variables corresponding to the position and velocity of the ball and racket. Each agent receives its own, local observation. Two continuous actions are available, corresponding to movement toward (or away from) the net, and jumping.
The task is episodic, and in order to solve the environment, your agents must get an average score of +0.5 (over 100 consecutive episodes, after taking the maximum over both agents). Specifically,
- After each episode, we add up the rewards that each agent received (without discounting), to get a score for each agent. This yields 2 (potentially different) scores. We then take the maximum of these 2 scores.
- This yields a single **score** for each episode.
The environment is considered solved, when the average (over 100 episodes) of those **scores** is at least +0.5.
## Getting started
### Prerequisites
- Python 3.6 (or more recent)
- [pip](https://pip.pypa.io/en/stable/)
- [ml-agents](https://github.com/Unity-Technologies/ml-agents) v0.4 (check this [release](https://github.com/Unity-Technologies/ml-agents/releases/tag/0.4.0b) if you encounter issues)
### Installation
You can install the project requirements as follows:
```shell
git clone https://github.com/frgfm/drlnd-p3-collaboration-competition.git
cd drlnd-p3-collaboration-competition
pip install -r requirements.txt
```
Download the environment build corresponding to your OS
- Linux: [here](https://s3-us-west-1.amazonaws.com/udacity-drlnd/P3/Tennis/Tennis_Linux.zip)
- Mac OSX: [here](https://s3-us-west-1.amazonaws.com/udacity-drlnd/P3/Tennis/Tennis.app.zip)
- Windows (32-bit): [here](https://s3-us-west-1.amazonaws.com/udacity-drlnd/P3/Tennis/Tennis_Windows_x86.zip)
- Windows (64-bit): [here](https://s3-us-west-1.amazonaws.com/udacity-drlnd/P3/Tennis/Tennis_Windows_x86_64.zip)
Then extract the archive in the project folder.
If you wish to use the agent trained by repo owner, you can download the model parameters as follows:
```shell
wget https://github.com/frgfm/drlnd-p3-collaboration-competition/releases/download/v0.1.0/maddpg_actor.pt
```
## Usage
### Training
All training arguments can be found using the `--help` flag:
```shell
python train.py --help
```
Below you can find an example to train your agent:
```shell
python train.py --deterministic --no-graphics
```
### Evaluation
You can use an existing model's checkpoint to evaluate your agent as follows:
```shell
python evaluate.py --checkpoint ./maddpg_actor.pt
```
## Credits
This implementation is vastly based on the following papers:
- [Asynchronous Actor Critic](https://arxiv.org/pdf/1602.01783.pdf)
- [Proximal Policy Optimization](https://arxiv.org/pdf/1707.06347.pdf)
- [DDPG](https://openreview.net/pdf?id=SyZipzbCb)
## License
Distributed under the MIT License. See `LICENSE` for more information.