piSTAR Lab

WARNING: This is an early release.

Overview

piSTAR Lab is a modular deep reinforcement learning platform built to make AI experimentation accessible and fun.

Documentation https://pistarlab.readthedocs.io

Features

• Web UI
• Extension System for adding new agents, environments or tasks types
• Python API, anthing you can do in the UI, you can do in Python as well
• Run agents in single and multi player environments
• Experiment tracking
• Uses Ray Project (https://ray.io/) under the hood for distributed processing
• Includes piSTAR Landia a hackable Multi Agent Envrionment
• More to come

Known Issues/Limitations

• Cluster mode is under development and not recommended at this time
• Running remotely requires SSH tunneling. All services must be running on localhost
• Mac not tested

UI Screenshots

Quick Start

More detailed documentation is available at https://pistarlab.readthedocs.io

Requirements

• Ubuntu suggested but also tested on Windows 10. We suggest using Docker for other setups.
• Python 3.7 or 3.8 (Conda is recommended)
  • Miniconda
  • Anaconda
• FFMPEG (Optional)
  • Required for episode recordings
  • Windows install instructions
• Xvfb (Optional, Ubuntu Only)
  • Helps some environments run without opening a window.
  • Useful when running piSTAR Lab remotely

Installation

For non-standard installations see: https://pistarlab.readthedocs.io/en/latest/installation.html

Create and Activate Conda Virtual Environment

conda create -n pistarlab python=3.7
conda activate pistarlab
conda install pip

Install with pip

pip install https://github.com/pistarlab/pistarlab/archive/refs/heads/main.zip#egg=pistarlab[all]

Usage

To launch piSTAR Lab UI, run:

pistarlab_launcher

Open browser to: http://localhost:7777

Contributing

We are still in an early phase of this release but if you are interested in contributing to piSTAR Lab, please reach out.