tmjeong1103 / rl_with_ray

RL_with_RAY

Implementation of reinforcement learning algorithms that is applied RAY in PyTorch.

In this repo, we implemented this RL_with_RAY code only use RAY CORE from scratch, didn't use RLLib.

Dependencies

1. Python 3.6.x
2. PyTorch 1.8.1
3. RAY 1.3.0
4. pybullet-gym
5. numpy

Install Requirements

pip install -r requirements.txt

Contents

PPO with RAY

Hyperparameter

SAC with RAY

Hyperparameter

ARS with RAY

Hyperparameter

- n_cpu = n_workers = 5
- hdims = [256,256]
- actv = nn.Tanh()
- out_actv = nn.Tanh()

Run

Set configuration

1. n_worker
2. total_steps
3. ep_len_rollout
4. hdims

Reference

1. https://docs.ray.io/en/master/index.html, RAY Documentation
2. https://arxiv.org/abs/1707.06347, PPO Paper
3. https://arxiv.org/abs/1801.01290, SAC Paper
4. https://arxiv.org/abs/1803.07055, ARS Paper