This is the page about the work done for my Bachelor thesis in Mathematics at the University of Turin.
MCTS algorithm has been used with success for games like Go, simulating a number of games randomly and creating a game tree. It can be enhanced further with UCT, a move selection strategy making a trade-off between exploration of new strategies and exploitation of better ones.
In this thesis, the algorithm working principle is explained. Moreover, it is discussed how it may be applied successfully also to videogames and an applied use is shown.
The game chosen as application is Pokémon Contest Spectacular, in which four players challenge themselves in a five-turn exhibition. Each turn the participants make a move to appeal the public or to jam opponents. Being a turn-based game with simultaneous moves and random elements, this is useful to show MCTS efficiency even outside turn-based sequential deterministic games such as Go.
Finally, it is analysed the efficiency of the algorithm, comparing its performance against random players, MCTS players of different levels (varying depending on the number of simulations performed), original game AI player and human players.
Here are some documents about the thesis. Please note that all the material is in Italian.
It is possible to download the software made for this thesis in different versions. Please note that all the material is in Italian.
contest.exefile, which will run the software.
contest.lovefile, which however requires the setup of the LÖVE framework on the computer. LÖVE works on Windows, Mac OS X and Linux and allows this software to be run on all these platforms.