Search results
30 gru 2019 · A reinforcement learning approach to the game 2048. The second post in a 3-part series dedicated to playing 2048 with AI. Part 1 (the game itself) available here, and part 3 (Monte Carlo tree search) available here. Series recap; Formulation. Choosing moves; Finding \(Y\) Loss. Weighting loss. Everything’s relative; Weighting. Defining ...
12 mar 2014 · With just 100 runs (i.e in memory games) per move, the AI achieves the 2048 tile 80% of the times and the 4096 tile 50% of the times. Using 10000 runs gets the 2048 tile 100%, 70% for 4096 tile, and about 1% for the 8192 tile.
Watch the AI join the numbers and get to the highest tile it can!
Join numbers to get to the 2048 tile! How to play 2048: Use arrow keys to move the tiles. When two tiles having the same number touch, they join into one!
How to play: Use your arrow keys to move the tiles. When two tiles with the same number touch, they merge into one! Note: This site is not the official version of 2048. It's an AI written for the game. The AI uses expectimax search to find the best move for the current board. Created by Gabriele Cirulli.
AI for the game 2048. See it in action here. (Hit the auto-run button to let the AI attempt to solve it by itself) The algorithm is iterative deepening depth first alpha-beta search.
AI for the 2048 game. This uses expectimax optimization, along with a highly-efficient bitboard representation to search upwards of 10 million moves per second on recent hardware. Heuristics used include bonuses for empty squares and bonuses for placing large values near edges and corners. Read more about the algorithm on the StackOverflow answer.
