CNN vs NN for Reinforcement Learning Flappy Bird

This project was an extension of the Reinforcement Learning Flappy Bird Project and aimed to look at the performance differences between using a CNN and a Neural Network for the task.

Using a CNN with frame stacking offered a nearly 10x increase in the performance, able to achieve 10s of thousands of timestamps being colliding!

I wrote a blog about this project here: 

Blog Entry

The full code can be found here:

GitHub Link