Seeing all the cool game AI's that were able to beat human players.
What it does
This project creates a scalable AI platform that can be used to play Tetris better than any human player. With further modifications, our model could also be used for other applications/games as long as it is fed the necessary parameters.
How we built it
We built this project using Python3 as our language of choice. We used PyGame for the frontend as well as native Python with some automation from NEAT to help expedite the training process of our neural network.
Challenges we ran into
We had some slight issues along the way namely, getting the Tetris logic functioning properly as well as some challenges understanding some of the docs for libraries we used.
Accomplishments that we're proud of
We were able to successfully complete this project as well as see meaningful results from our AI.
What we learned
We learned how to implement feed-forward neural networks as well as further understand biases in data and explore different models by weighing their benefits and downsides (for example higher risks of overfitting/underfitting by a specific algorithm such as kNN).
What's next for NEATris
Implementing more awareness in our neural net including implementing T-Spins, as well as probabilistic predictions that would help the AI prioritise potential future moves that would bring higher scores. We would also like to carry this project over to different games and see if we can get it to train and run in different environments like Mario Kart or Pacman.