subeans / Deep-rl-mxnet

Mxnet implementation of Deep Reinforcement Learning papers, such as DQN, PG, DDPG, PPO

Deep-rl-mxnet

Mxnet implementation of Deep Reinforcement Learning papers.

Now this repository contains:

• DQN (Simple implementation) -->MountainCar-v0
• Double DQN (Simple implementation) -->MountainCar-v0
• Dueling DQN (Simple implementation) -->MountainCar-v0
• Policy Gradient (Simple implementation) -->CartPole-v0
• DDPG (Detailed implementation) ⭐ -->Pendulum-v0,LunarLander-v2
• PPO (Simple implementation) -->CartPole-v0
• TD3 (Very detailed implementation) ⭐ ⭐ -->Pendulum-v0,LunarLander-v2,HalfCheetah-v2

In the future this will contain:

1. A3C
2. SAC

Requirements:

• Python 3
• OpenAI gym
• Mxnet(gpu or cpu) and gluonbook
• Box2d(optional)
• Mujoco(optional)

Basic Installation:

pip install gym
pip install gluonbook
pip install mxnet  (cpu version)
pip install mxnet-cu90  (gpu version, corresponding to your cuda version)

Box2d Installation(Optional):

pip install box2d.py

If you get something like this:

unable to execute 'swig': No such file or directory

do:

sudo apt-get install swig

Mujoco Installation(Optional):

Please refer to this repository

Contents:

1. DQN

• Playing Atari with Deep Reinforcement Learning
• Human-level control through deep reinforcement learning
  

2. Double DQN

• Deep Reinforcement Learning with Double Q-learning

3. Dueling DQN

• Dueling Network Architectures for Deep Reinforcement Learning

4. Policy Gradient

• Policy Gradient Methods for Reinforcement Learning with Function Approximation

5. Deep Deterministic Policy Gradient

• Continuous Control with Deep Reinforcement Learning

6. Proximal Policy Optimization

• Proximal Policy Optimization Algorithms
• Emergence of Locomotion Behaviours in Rich Environments

7. TD3

• Addressing Function Approximation Error in Actor-Critic Methods

8. A3C

• Asynchronous Methods for Deep Reinforcement Learning

9. SAC

• Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor