PyTorch implementation of Deep Reinforcement Learning Algorithms

This repository contains :

1. Value-Based Methods : (Neural Q-Learning, DQN, Double-DQN, Memory improved DQN)
2. Policy-Based Methods: (DDPG)

Important notes

• The code works for PyTorch.
• The agents interact with OpenAI gym and Unity environments.

Features

• Support CUDA.(Faster than CPU implementation)
• Support discrete and continous state space.
• Support discrete and continous action space.

Resources

• Deep RL Bootcamp
• Deep Reinforcement Learning UC Berkeley
• Udacity Deep Reinforcement Learning Nanodegree program

OpenAI Gym Benchmarks

Classic Control

• Acrobot-v1 with Coming soon!
• Cartpole-v0 with REINFORCE | solved in 691 episodes.
• MountainCarContinuous-v0 with DDPG
• MountainCar-v0 with Coming soon!
• Pendulum-v0 with DDPG

Box2d

• BipedalWalker-v2 with DDPG
• CarRacing-v0 with Coming soon!
• LunarLander-v2 with NeuralQLearner | solved in 314 episodes. Average Score: 200.5

Toy Text

• FrozenLake-v0 with Coming soon!
• Blackjack-v0 with Coming soon!
• CliffWalking-v0 with Coming soon!

Unity-ML-Agents Benchmarks

• BananaCollector with NeuralQLearner | solved in 345 episodes. Average Score: 13.02
• Reacher with DDPG | solved in 147 episodes. Average Score: 30
• Tennis with MADDPG | solved in 427 episodes. Average Score: 0.5

Linux dependencies

To set up your python environment to run the code in this repository, follow the instructions below.

1. Install conda and create a new environment with Python 3.6.

conda create --name drlenv python=3.6

2. To activate this environment

source activate drlenv

3. To install and use OpenAI gym.

• Next, install the classic control environment group by following the instructions here.
• Then, install the box2d environment group by following the instructions here.

4. [Optional] To install and use Unity ML-Agents

5. Clone the repository

git clone https://github.com/dganbold/deep_reinforcement_learning

6. Install the dependencies

pip install .

References

• Udacity Deep-Reinforcement-Learning [Github]