Reversi Game AI Implementation

AI Midterm Project, our group implement minimax with alpha-beta pruning and Monte Carlo Tree Search these to algorithms

Transcript

1. Reversi AI Implementation - Dante Lin - Pokai Chang - Yukai Huang
2. Game Logic
3. http://omerkel.github.io/UCThello/html5/src/ (獨Ո狶ጱ)
4. Algorithms
5. Minimax α–β pruning & Monte Carlo Tree Search
6. System Design
7. JS + HTML Canvas
8. Pluggable AI Agent
9. Directory Structure
10. src/ !"" DrawBoard.js !"" Game.js !"" ReversiBoard.js !"" assets # !"" images # !"" index.html # !"" minimax.html # $"" minimax.js !"" player # !"" BasePlayer.js # !"" HumanPlay.js # !"" MctsPlayer.js # !"" RandomPlay.js # !"" index.js # $"" mcts !"" stylesheets # $"" style.css $"" utils $"" index.js
11. AI Agents Show me the code
12. export default class RandomPlay extends BasePlayer { getMove(board) { var validPositions = board.getValidPositions(); return( validPositions[ this.randomPick(0, validPositions.length-1) ]); } randomPick(min, max) { return Math.floor(Math.random()*(max-min+1)+min); } } Random
13. monteCarloTreeSearch(board) { var rootNode = new Node(board); var simCount = 0 var begin = new Date().getTime(); while (new Date().getTime() - begin < 3000) { var pickedNode = this.treePolicy(rootNode); var result = this.simulate(pickedNode.gameState); this.backPropogation(pickedNode, result); simCount += 1; } return this.bestAction(rootNode) } MCTS
14. bestChild(node) { var C = 1; // 'exploration' value var max = { node: null, value: Number.NEGATIVE_INFINITY }; for (var child in node.children) { var { wins, plays } = node.children[child].getWinsPlays(); var { plays: parentPlays } = node.getWinsPlays(); var value = (wins / plays) + C * Math.sqrt(2 * Math.log(parentPlays) / plays); if ( value > max.value ) { max.node = node.children[child]; max.value = value; } } return max.node; } MCTS (cont.)
15. Minimax var boardData = [ _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, O, X, _, _, _, _, _, _, X, O, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, ];
16. var scoreBoard = [ 90, -60,10,10,10,10,-60, 90, -60, -80, 5, 5, 5, 5,-80,-60, 10, 5, 1, 1, 1, 1, 5, 10, 10, 5, 1, 1, 1, 1, 5, 10, 10, 5, 1, 1, 1, 1, 5, 10, 10, 5, 1, 1, 1, 1, 5, 10, -60, -80, 5, 5, 5, 5,-80,-60, 90, -60,10,10,10,10,-60, 90 ]; Minimax (cont.)
17. Minimax (cont.) function evaluate(color, board){ ... for(var i = 11; i < 89; i++){ if(board[i] == color){ boardS += scoreBoard[i-11]; } } count_score = m_weight*mobility + b_weight*boardS; return count_score; }
18. Minimax (cont.) function minmax(turn, depth, alpha, beta, board){ ... if(!turn){ var value=-9999; for(i = 11; i < 89; i++){ if(isLegal(1,2,i)){ touchMove(1,2,i,c_board); score=minmax(true,depth-1,alpha,beta,c_board); if(score>value){ value=score; } alpha=Math.max(alpha,score); if(alpha>=beta){ break; } } } } ...
19. if(turn){ var value=9999 for(i=11;i<89;i++){ if(isLegal(2,1,i)){ touchMove(2,1,i,c_board); score = minmax(false,depth-1,alpha,beta,c_board); if(score < value){ value=score; } alpha=Math.min(beta,score); if(alpha<=beta){ break; } } } } return value; } Minimax (cont.)
20. function moveWithMinmax(){ var value=-9999; var score; var pos; for(var i=11;i<89;i++){ if(isLegal(1,2,i)){ score=minmax(false,depth,-9999,9999,testBoard); if(score>value){ value=score; pos=i; } } } return pos; } Minimax (cont.)
21. Live Demo http://goo.gl/Juiq2l