# LR-RL

**Repository Path**: shylockH/lr-rl

## Basic Information

- **Project Name**: LR-RL
- **Description**: somthing to learn Reinforcement Learning
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2022-10-02
- **Last Updated**: 2023-01-06

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# A basic framework to Learn RL
## Introduction
This is a framework to learn reinforcement learning. Some basic code is provided in directory **./util**, you may use it to help you implement RL algorithm. 
## To implement algorithms
A RL algorithm can be abstracted into two parts : an agent to update its policy by a batch data and a trainer to sample data from environments by a certain policy. A friendly way to implement a RL algorithm with this framework it to inherit the base class ***BaseTrainer*** in **./util/trainers** and ***BaseAgent*** in **./util/agents**. Then implement some abstractmethods of them.
## To train an agent
Some basic RL algorithms, like DQN, DDPG, SAC and so on, are also implemented in **./algo**. For sac algorithm, just use it directly like this:  
``
python ./algo/sac/main.py --task Halfcheetah-v3
``  
You could adjust the hyperparameters in **./algo/sac/config.py**