重点看 Pendulum/A3C

对torch.multiprocessing 讲解 及A3C的实现 通俗易懂

Introduction

Using (deep) reinforcement_learning algorithm to practice on OpenAI Gym, Unity ML-Agents,and other virtual environments.

For more information about OpenAI Gym environment, see: https://github.com/openai/gym

Index

1. Acrobot

A Reinforcement Learning project running on OpenAI gym environment: Acrobot-v1

Using tile coding to solve the continuous state space, and them perform the original Q-Learning algorithm to it.

2. MountCar

A Reinforcement Learning project running on OpenAI gym enviroment: MountainCarContinuous v0.

Using discretization to generalize the Q-Learning algorithm to continuous space.

3. Pro_BlackJack

An implementation of Monte Carlo controlling algorithms to solve the OpenAI gym environment:Blackjack*-v0.

4.Pro_LunarLander

An implementation of deep reinforcement learning algorithm—DQN， except for the original DQN form, I also tried several improved architecture, including double DQN, Prioritized Experience Replay, and the dueling DQN. I use Pytorch to build the neural network for my agent.

the environment I use to experiment the DQN agent is LunarLander-v2

5. MountCar_continuous

Using mutiple approaches to solve the gym MountainCarContinuous v0 https://github.com/openai/gym/wiki/MountainCarContinuous-v0

• cross_entropy_method