Rishabh Prabhu

How can you use data science to analyze the outcomes in a turn based scenario with random components?

Started Apr. 11, 2022

Abstract or project description

The first part is acquiring and parsing through data. Then you have to evaluate the data and see if it is possible for the machine to adapt to scenarios dependent on luck. The data we will get will be from Pokémon as it is a game that you can learn and grow as a player through playing the game, but has luck based components that can lead to lots of different outcomes, which separates ML playing Pokémon from ML playing most board games, as most board games have zero luck dependency. Pokémon has a pretty simple set of moves with at most 9 options per turn for 2 players. The main focus will to be to analyze what options are chosen in specific scenarios and to see if said options leads to wins or losses, and to analyze the effectiveness of said options using a calculator. Data for Pokémon can be extracted from Pokémon showdown, and the GitHub, and via match transcripts from match replays that will be the main data source. The random components will change the scenario from predicting to adapting to see the capabilities of ML to react rather than predict, as in certain cases predicting all the options can require a lot of computing power, which might not be available. Also we will be able to see how the ML actually chooses to gamble on luck rolls and see if that correlates to a 50/50 or if the ML either chooses more consistent rolls, or gamble more often on lower probability rolls.