Source: Deep Learning on Medium
NEAT Game: how I complete a project
NEAT Story Part 3
Day 0 Hour 0: I’m going to make a simple game to show off my implementation of NEAT in a tangible way. Just talking about neural architecture is very abstract. I easily visualize the members of each generation performing crossover and generating neural connection mutations at a .03% rate, and new neurons at a .02% rate. I have also been immersed in this research paper for 10 months. While I might not be able to animate this example, I can at least endeavor to make some geometric creatures spawn and compete for resources. An idea inspired by the work of my friend Tatsuaki Ilya. I had a few conversations with him about it before starting my own spin off to demonstrate NEAT.
I’ve never animated anything or made a game in Python. I know there are several options for graphics and games, but I’ve already decided to start by trying Turtle a Tkinter extension. I’m going to try to finish this in 5 hours, so I know I need to plan for 10. I just don’t have a lot of time for another side project in the next few weeks. I’ve watched one 6 minute tutorial on the topic. https://youtu.be/PTgyzZGknvg
First I’ll map out my hourly goals, starting from the finish line.
Hour 5: Finish debugging game, and polishing graphics
Hour 4: Ensure OpenAI Gym format works with my implementation of NEAT
Hour 3: Modify code into the format of an OpenAI Gym environment
Hour 2: Finish making game eating, interaction and starvation
Hour 1: Start making game graphics and movement
This seems reasonable. I’m not going to describe in too much detail the game yet. I want to remain flexible so that I can quickly pivot to the simplest solutions. If I remain tied to a particular idea about how it “has” to be I could loose my entire time frame trying to solve a small problem that turns into a bottomless pit of Stack Overflow and Python documentation searches.
Day 1: Hour 1: I’m still sure that my goal of 5 hours is too ambitious, but I’m going for it anyway. Wish me luck, or at least focus.