Home

SimionZoo

This project features a set of tools/applications written in either C++ or C# designed to make experiments with Reinforcement Learning algorithms on control tasks with continuous state and action spaces. The main goal is to provide an easy-to-use environment in which end-users (no programming skills required) can design, run, and monitor/view experiments, and then analyze the results. The most prominent features are:

• Experiment parameters can be given a set of values to perform a parameter sweep
• All the different combinations can be run in parallel using the built-in distributed execution mode
• The results of an experiment can be analyzed with customizable plots
• The behavior of system can also be viewed live or after an experiment has finished
• It supports Windows (x86 and x64) and Linux (x64).

Installation

End-users who want to run the binaries should read this guide. Developers who want to compile the sources should read this guide.

Main applications:

• RLSimion: a C++ console application that does the actual Reinforcement Learning, taking an XML file as input and generating a log as output.
• Badger: a C# WPF GUI tool that allows to design experiments and run them locally or in remote machines.
• Herd agent: a C# Windows service that needs to be installed in those machines intended to work as slaves from Badger.

Acknowledgements

The code features contributions from:

• Unai Tercero (Badger and Herd Agent)
• Asier Rodríguez (Bullet worlds)
• Alejandro Guerra (Badger and Herd Agent)
• Roland Zimmermann (Badger, OffPAC, INAC, Tile Coding, ... and all about CNTK and Deep RL)

Except a few fixes by others (check the commmits), the rest of the code has been written and is mantained by Borja Fernández-Gauna from the Group of Computational Intelligence at the University of the Basque Country (UPV/EHU), so any questions/suggestions can be directed to my email address: borja.fernandez'at'ehu.eus.