Using Neural Networks to Discover new Tetris™ Strategies
Faculty Sponsor
Tiffany Kolba
College
Arts and Sciences
Discipline(s)
Data Science
Presentation Type
Oral Presentation
Symposium Date
Spring 4-29-2021
Abstract
Using neural networks to play Tetris is a classical implementation. Tetris is a game where there are immediate solutions to gain points, but there are also less obvious choices that a player can choose to have delayed benefit, such as adding a gap and waiting for an I-piece. This paper presents a new heuristic added to the repertoire of classically implemented Tetris heuristics, with the goal of making discoveries in a technique known as bagging. Tetris generates a bag of the 7 tetrominos {I, O, T, J, L, S, Z} that are then chosen and removed from the bag as the pieces drop. Once the bag is empty, the game generates a new one. Tetris players take advantage of this fact to clear the board using several bags, with no pieces left over, letting the players have an infinitely repeatable strategy. The difficulty in this is finding patterns that can be repeated regardless of piece order. Once our heuristic was implemented, our algorithm began frequently repeating moves and managed to replicate some bagging behavior. However, due to these repeated moves the neural network was no longer able to play for extended periods of time, likely due to new unseen bag orders.
Recommended Citation
Randle, Nathan, "Using Neural Networks to Discover new Tetris™ Strategies" (2021). Symposium on Undergraduate Research and Creative Expression (SOURCE). 975.
https://scholar.valpo.edu/cus/975
Biographical Information about Author(s)
Nathan Randle, Data Science Major, Expected 2021