This repository will consist fast 2048 that can be played in console and few AI algorithms that can be compered against each other.
Now it contains working game in console (without winning conditions) nn branch with basic neural network definitions in progress
- Add all neural components
- Add scoring system for the game
- Add game condition to win and lose game
- Add fitness function
- Random shuffle
- Random shuffle only going up and both sides (can be done in minutes)
- Neural network with evolution trainning
- Neural network with Q learning
- MinMax with different predicted number of moves
- Expectiminimax
- Basic hardcode algorithm (to compare)
This approach takes random number to select move and preforms it. With added additional preffered direction algorithm will try to push all cells to one side to maximize chance of winning (simple strategy used by players)
Well it is kinda self explanatory. In this particular case array of whole game board is given with score and the output is array of moves.
Find best move by looking at possible outcomes (simulate x next game states)
Really similar to MinMax but added few corrections to uncertainties due to random values.
Works similar to random shuffle with limited one way. This algorithm will try to do some basic stuff to follow zig zag pattern starting from top and slowly descending in order.