2048 AI

AI Plays http://gabrielecirulli.github.io/2048/

Just for my term project...

Browser game control part is wholly from https://github.com/nneonneo/2048-ai, also idea of optimization techniques. (But we didn't use c, so still slow)

Should install Remote Control for Firefox

Not going to implement current state of the art methods (tree search with heuristic evaluation: https://github.com/nneonneo/2048-ai, temporal difference learning: https://github.com/aszczepanski/2048)

At first, MCTS was also considered but I suspended that plan because 2048 needs to be played very fast and random tile addition creates wide possibilities.

So just n-ply model of DQN(deep q learning or maybe value learning) is my plan

Currently implemented: random, maximum merge(greedy), monte carlo simulation, value learning (n-ply playing)