{"id":5243,"date":"2014-04-04T14:05:11","date_gmt":"2014-04-04T18:05:11","guid":{"rendered":"https:\/\/blogs.mathworks.com\/pick\/?p=5243"},"modified":"2014-04-04T14:09:42","modified_gmt":"2014-04-04T18:09:42","slug":"submit-your-algorithms-to-solve-2048","status":"publish","type":"post","link":"https:\/\/blogs.mathworks.com\/pick\/2014\/04\/04\/submit-your-algorithms-to-solve-2048\/","title":{"rendered":"Submit your algorithms to solve 2048!"},"content":{"rendered":"
\r\n

Jiro<\/a>'s pick this week is 2048 MATLAB<\/a>.\r\n <\/p>\r\n

\"2048\"<\/b><\/p>\r\n

Need I say more? I'm certain that many of you know what I'm talking about. 2048<\/a> is an online and mobile game created by Gabriele Cirulli recently. All I can say is that it is an addictive game that has\r\n already resulted in a few MATLAB implementations<\/a>.\r\n <\/p>\r\n

A couple of weekends ago, I also spent a day implementing a MATLAB version. I probably spent more time than I needed to on\r\n trying to make it look as close as possible to the original version, but it was a lot of fun.\r\n <\/p>\r\n

<\/p>\r\n

But then, one of my colleagues, Seth<\/a> (who was also just featured<\/a> in this blog last week), suggested including a mechanism for feeding in different algorithms to solve the game. There must\r\n be an AI algorithm that would solve the game or produce a high score consistently.\r\n <\/p>\r\n

In the app, I've included the ability to select a MATLAB function as the solving algorithm. Then you can see the algorithm\r\n in action as it tries to solve the game. The algorithm must be defined as a MATLAB function that takes in a 4x4 matrix representing\r\n the game board (NaN for empty spots) and returns a character array representing the direction to move ('up', 'down', 'right',\r\n or 'left'). Here's an example of a simple algorithm that randomly chooses a direction.\r\n <\/p>

function<\/span> direction = myAI(board)\r\n  d = {'up'<\/span>, 'down'<\/span>, 'right'<\/span>, 'left'<\/span>};\r\n  direction = d{randi(4)};<\/pre>
I've also included a command line simulator, which you can use to do a montecarlo simulation of a particular algorithm.<\/p>
game = Game2048Simulator(@myAI);\r\nsimulate(game, 1000)   % 1000 simulations<\/span>\r\nviewResult(game, 25)   % 25 histogram bins<\/span><\/pre>
 <\/p>\r\n   
We can display the highest score and compute what percentages of the simulation resulted in each of the highest block.<\/p>
% Display highest score<\/span>\r\ndisp(['Highest score: '<\/span>, num2str(max(game.Result.Score))])\r\ndisp(' '<\/span>)\r\n\r\n% Get the unique values for high blocks<\/span>\r\nhighblocks = game.Result.HighestBlock;\r\nblocks = unique(highblocks);\r\n\r\n% Compute the percentage of occurrance for each high block value<\/span>\r\nnSims = length(highblocks);\r\nfor<\/span> id = 1:length(blocks)\r\n    percentage = nnz(highblocks == blocks(id)) \/ nSims;\r\n    disp([num2str(blocks(id)), '   '<\/span>, num2str(percentage*100), '%'<\/span>])\r\nend<\/span><\/pre>
Highest score: 3172\r\n \r\n16   0.2%\r\n32   6.8%\r\n64   34.5%\r\n128   49.7%\r\n256   8.8%\r\n<\/pre>
Submit your algorithm!<\/b><\/p>\r\n   
So here's a challenge for you. Submit<\/a> your algorithm and win some MathWorks swag! I will give prizes for each of the following criteria:\r\n   <\/p>\r\n   
\r\n