Homegrown AI Connect 4

This project was my capstone during my Data Science training at General Assembly.  I built a Connect-4 (C4) game engine to see if I could train an Artificial Neural Network (ANN) to make decisions without explicitly stating the rules of the game. After some trial and error, I found that Convolutional Neural Networks (CNN) work best to recognize the winner of a game presumably because the grid can be considered the same as an image.  The model was trained by playing against an AI that places pieces at random over the course of 300,000 games. Eventually the CNN is able to defeat the Random placement 45% of the time. The model could be improved further by playing against itself, or other AI implementations. 

Make sure to check out the project on GitHub to read all about it ! 

Tags : Machine Learning, Neural Networks, CNN, Game Development, AI