Reinforcement_learning_with_pytorch

Implement some algorithms of RL

Pytorch version: 1.8.1+cudnn10.1

Requierment

• gym
• numpy
• pytorch: 1.8.1+cudnn10.1
• tensorboard

Implemented algorithms

Bandit algorithms

• UCB
• LinUCB

Model-free algorithms

• REINFORCE
• A2C(Advantage Actor-Critic)
• A3C
• DQN
• DoubleDQN
• DuelingDQN
• D3QN(DuelingDoubleDQN)
• DDPG
• PPO
• SAC
• SAC_Discrete

Model-based algorithms

• Dyna-Q
• MBPO
• PETS

Causal RL algorithms

to be continue...

How to run

Discrete action environment

We use Cartpole-v1 as our test environment.

python train_cartpole.py -a A2C

Continuous action environment

We use Pendulum-v1 as our test environment.

not yet completed...

Reference

• https://github.com/MorvanZhou/Reinforcement-learning-with-tensorflow
• https://github.com/boyu-ai/Hands-on-RL