Threes-MCTS

Threes! AI with Monte-Carlo Tree Search. This project is for Taiwan 2048 Bot competition. In this project, the threes! agent uses the monte-carlo tree search with UCT as the first trial. In the meantime, the other member is implementing a expeciminimax algorithm which has outperformed this MCTS algorithm.

Under the time constriant, the number of simulation is way too small.

To-do list

• Accelarate the simulation process
• Improve the original UCT algorithm
• Unit testing for each component

File Description:

Only covers the .h header files, but include coresponsing .cpp files

• Threes.h are provided by the competition organizer
• myrand.h, statistics.h and mt19937ar.h are from Dr. Yu
• GameAgent.h: interface of search algorithm
• GameAgentGame.h: class Game for simulation
• MonteCarloTreeSearch.h: the general framework
• MonteCarloTreeSearchNode.h: how MCTS actually works, UCT values are calculted here

Built With

• c
• c++