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...