claudecoulombe / evolutionmve

Minimum viable experiments using Reinforcement Learning QLearning and evolution strategy to play catch - by Eder Santana http://edersantana.github.io - Small corrections and addition of iPython Notebooks by Claude COULOMBE - PhD candidate - TÉLUQ / UQAM - Montréal

Code for Playing Catch with Keras and an Evolution Strategy blog post

Small corrections and addition of a iPython Notebook by Claude COULOMBE - PhD candidate - TELUQ / UQAM - Montreal

Train

python evolve.py

Test

1. Generate figures

python test.py

2. Make gif

ffmpeg -i %04d.png output.gif -vf fps=1

Requirements

• Prior supervised learning and Keras knowledge
• Python science stack (numpy, scipy, matplotlib) - Install Anaconda!
• Tensorflow (last test on version 1.4)
• ffmpeg (optional)

License

This code is released under MIT license.