pistarlab 0.0.1.dev0

A modular AI agent experimentation tool.

piSTAR Lab

WARNING: Under active development - not ready for public use.

Overview

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

Release: 0.0.1-dev (early release)

Features:

• Web UI
• Plugin 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
• Built in web-based IDE (via Theia (https://theia-ide.org/))
• Uses Ray Project (https://ray.io/) under the hood for distributed processing

Licensed under an Apache-2.0 license.

UI Screenshots

Quick Start

These instructions are for single node only. For cluster mode, see TODO

Installation with Anaconda

NOTE: Only tested on Ubuntu, but should also work on OS X. MS Windows users see Installation using Docker

1. Install Anaconda or Miniconda Visit https://www.anaconda.com/products/individual for instructions

2. Install PIP

   conda install pip
   

3. Clone Repo and install

   git clone https://github.com/pistarlab/pistarlab
   cd pistarlab
   pip install -e .
   

4. build Redis

   bash ./install_redis.sh_
   

5. install node for UI and IDE

   bash ./install_node.sh
   bash ./build_ui.sh
   bash ./build_ide.sh #optional
   

6. install additional dependencies

   • XVFB to render without display (No MS Windows Support)
   • ffmpeg for video processing

   sudo apt-get install -y xvfb ffmpeg
   

Usage

To launching piSTAR Lab, run:

pistarlab

• UI: http://localhost:8080

• Launcher Control Panel: http://localhost:7776

Installation using Docker

Recommended for MS Windows users

1. Install Docker:

   Visit https://docs.docker.com/engine/install/

2. Clone Repo

   git clone https://github.com/pistarlab/pistarlab
   cd pistarlab
   

3. Build Docker Image

   ./build_docker.sh
   

Usage with Docker

Launching piSTAR Lab

./bin/docker_launcher.sh 

Contributing

We are still in an early phase of this release and have many loose ends to wrap up. If you are interested in contributing to piSTAR Lab, please reach out.