Monte Carlo Tree Search Implementation

Somewhat general implementation of Monte Carlo Tree Search. Only for games that the rewards are all provided at the termination of the game (could easily be adapted though). Note: final rewards should have |r| <= 1. Need to provide a number of functions to define the game logic. I made this mainly to get some practice using the c++ standard template library and as such have made no attempt to ensure that the code is particularly efficient. There are likely to be much faster implementations of the algorithm out there!

Main.cpp has an example of using the template with a simple implementation of a tic tac toe game.

What needs to be provided:

• A class for the game state. should include an integer denoting players go (0,1,...) in variable plyrsGo
• A class for the possible actions. (probably easiest to use an integer in many cases).
• A function getActions(state) (returns vector of actions available)
• A function getNextState(state, int)
• A function defaultPolicy(state, vector actions)
• A function selectionEvaluation(double,int, int) that evaluates how we select "intelligently" (e.g. UCB1)
• A function isTerminal(state)
• A function assignRewards(state) that evaluates the terminal state and determines which player won.

e.g. usage:

currentGame Game(state);

Game.initialize;

Game.playNGames(N);

A moveToMake = Game.bestInitialAction();

Game.cleanUp();

Game.playNGames(N); //play again from the next state that's actually been chosen.