RLP: RL Algorithms Implemented by Pytorch

RLP is a set of deep reinforcement learning algorithms which implemented by Pytorch.
Just want to have a deeper understanding of the idea of these algorithms, and meanwhile provide some useful tools for others.
If you have some problems, please feel free to discuss.😁
Advantages: Easy to understand, Concise, Uniform code format
Notice: this implemented based on OpenAI Spinning Up and Others.

---

Done (but still need to be optimized)

Off-Policy:

• DDPG
• TD3
• SAC

On-Policy:

• VPG
• TRPO
• PPO

Continue Updating...

• DQN
• ...

---

Requirements

• gym
• mujoco-py
• PyTorch(1.0.1)
• Python(3.6)
• mpi4py

Run

Eg: For DDPG
python ddpg.py --env HalfCheetah-v2 ...(other parameters)

Test

Eg: For DDPG
python test_policy.py {ddpg model path} -num {choose a model} ...(other parameters)

Benchmarks

Mujoco