PARL

PARL is a flexible, distributed and object-oriented programming reinforcement learning framework.

Overview

• Features
• Abstractions
• Parallelization

Installation

• Installation

Tutorial

• Getting Started
• Model, Algorithm, Agent
• Create Customized Algorithms
• Save and Restore Parameters
• Visualization Tool
• CSV Logger

High-quality Implementations

• Policy Gradient
• DQN
• DDPG
• DDQN
• OAC
• A2C
• TD3
• QMIX
• SAC
• PPO
• MADDPG

Parallel Training

• Overview
  • Easy-to-use
  • High performance
  • Web UI for computation resources
  • Supporting various frameworks
  • Why PARL
• Xparl Usage
  • Setup Command
  • Adding More Resources
  • Example
  • Shutdown the Cluster
  • Further Reading
• Recommended Practice
• Recommended Practice(no_wait mode)
• GPU Cluster
• How to Debug
• File Distribution
  • Example:
• Serialization Acceleration (Not Necessary)
  • Performance Comparison
  • Conclusions

APIs

• parl.Model
• parl.Algorithm
• parl.Agent
• parl.remote_class
• parl.connect

Frequently Asked Questions

• xparl questions
• RL questions

EvoKit

• Overview
• minimal example
• Example for Online Products