# atari-rl

**Repository Path**: goforfar/atari-rl

## Basic Information

- **Project Name**: atari-rl
- **Description**: Atari - Deep Reinforcement Learning algorithms in TensorFlow
- **Primary Language**: Unknown
- **License**: MIT
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2020-06-30
- **Last Updated**: 2020-12-19

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# Atari - Deep Reinforcement Learning algorithms in TensorFlow

[![Build Status](https://travis-ci.org/brendanator/atari-rl.svg?branch=master)](https://travis-ci.org/brendanator/atari-rl)

Learning to play Atari in TensorFlow using Deep Reinforcement Learning

## Setup
```
git clone https://github.com/brendanator/atari-rl
git submodule update --init
conda create --name atari-rl python=3.5
source activate atari-rl
conda install -y -c https://conda.binstar.org/menpo opencv3
conda install -y h5py numpy
pip install tensorflow
pip install 'gym[atari]'
```

with without pip
```
virtualenv --system-site-packages -p python3 envName
pip install opencv-python
pip install h5py
pip install 'gym[atari]'
pip install numpy
pip install tensorflow
```
Python 2.7 is also supported

## Usage

- Show all options - `python main.py --help`
- Play a specific [Atari game](https://github.com/mgbellemare/Arcade-Learning-Environment/blob/master/src/games/Roms.cpp#L17) - `python main.py --game Breakout`

## Papers Implemented

- :white_check_mark: [Human Level Control through Deep Reinforcement Learning](http://www.nature.com/nature/journal/v518/n7540/pdf/nature14236.pdf)
    - `python main.py`
- :white_check_mark: [Deep Reinforcement Learning with Double Q-learning](https://arxiv.org/pdf/1509.06461.pdf)
    - `python main.py --double_q`
- :white_check_mark: [Dueling Network Architectures for Deep Reinforcement Learning](https://arxiv.org/pdf/1511.06581.pdf)
    - `python main.py --dueling`
- :white_check_mark: [Learning to Play in a Day: Faster Deep Reinforcement Learning by Optimality Tightening](https://arxiv.org/pdf/1611.01606.pdf)
    - `python main.py --optimality_tightening`
- :white_check_mark: [Prioritized Experience Replay](https://arxiv.org/pdf/1511.05952.pdf)
    - `python main.py --replay_prioritized`
    - Only proportional prioritized replay is implemented
- :white_check_mark: [Unifying Count-Based Exploration and Intrinsic Motivation](https://arxiv.org/pdf/1606.01868.pdf)
    - `python main.py --exploration_bonus`
- :white_check_mark: [Deep Exploration via Bootstrapped DQN](https://arxiv.org/pdf/1602.04621.pdf)
    - `python main.py --bootstrapped`
- :white_check_mark: [Increasing the Action Gap: New Operators for Reinforcement Learning](https://arxiv.org/pdf/1512.04860.pdf)
    - `python main.py --persistent_advantage_learning`
- :white_check_mark: [Learning values across many orders of magnitudes](https://arxiv.org/pdf/1602.07714.pdf)
    - `python main.py --reward_scaling`
- :white_check_mark: [Asynchronous Methods for Deep Reinforcement Learning](https://arxiv.org/pdf/1602.01783.pdf)
    - `python main.py --async one_step`
    - `python main.py --async n_step`
    - `python main.py --async n_step --sarsa`
    - `python main.py --async a3c`
- :x: [Deep Recurrent Q-Learning for Partially Observable MDPs](https://arxiv.org/pdf/1507.06527.pdf)
- :x: [Safe and efficient Off-Policy Reinforcement Learning](https://arxiv.org/pdf/1606.02647.pdf)
- :x: [Continuous Deep Q-Learning with Model-based Acceleration](https://arxiv.org/pdf/1603.00748.pdf)

## Acknowledgements

- https://github.com/mgbellemare/SkipCTS - Used in implementation of [Unifying Count-Based Exploration and Intrinsic Motivation](https://arxiv.org/pdf/1606.01868.pdf)
- https://github.com/Kaixhin/Atari
- https://github.com/carpedm20/deep-rl-tensorflow