Under construction. Check out the blog post on this!
This a project I have been working on for many years. I went into the details of this project on my blogs page. In summary, I developed a neural network to identify optimal characters to play in a competitive game by identifying key combinations of champions that synergize well while being good against the enemy champions.
Things I used and learned:
The problem itself is very difficult. Even if one team has a perfect team composition, those players can still be worse, or not be in the best condition, not focused, etc. With my best efforts, I was able to achieve an accuracy of 55% which seems to be the limit for this problem. I found commercial tools online that had basically the same accuracy with 10x the data I had.
The largest boost to my accuracy was switching to embedding layers. Eliminates specific champion trends and treats champions more like specific types of gameplay. This solved the overfitting issue.
You can read more on this on my blog page!
This is the project displayed on the home page. This was a larger endeavor than I thought initally. Some of the core things I learned/used in this project were:
Boids by itself fairly trivial, but an additional requirement I made was for the fish to A) have boundary logic, and B) also swim for food when the user feeds them. This made the project a bit more challenging. Also, you might have noticed if you refreshed the page a couple times, is that the koi fish color patterns are random. I did this by applying a 3d Simplex noise function to the fish with per instance random offsets in the vertex shader. Colorful (and unique) fish!
I also learned how to 3D model, and to create materials. I happily embraced the low-poly style since anything more would be way too difficult.
In '_banner_pixelated.js', I attempted to make a pixel-art styled renderer for this, but it didn't feel quite right so I scrapped it.
This is a project solving a website game called Fruit Box by GameSaien, and I explained this project on my Github.
Here is a short rundown of the game.
I spent days playing this game but the highest score I got was 109 out of 170. Can we do better?
In this project I learned many concepts.
Unfortunately, I couldn't cause an apple extinction...