Expectimax
A game-tree search algorithm and variant of Minimax designed for decision-making scenarios involving random events. It maximizes value at player nodes and computes expected values at chance nodes, making it suitable for games with stochastic elements such as 2048.
Timeline (last 90 days)
A developer attempted to build a 2048 game auto-solver using a genetic algorithm combined with Expectimax search; the current implementation can reach the 2048 tile in 30% to 50% of games.
Expectimax is more suitable than Minimax for 2048 because new tiles in 2048 are randomly generated and there is no opponent's will; Expectimax takes the maximum value at player decision nodes and the expected value at random generation nodes.
The role of genetic algorithms in 2048 solvers is typically to optimize the weights of each feature in the Expectimax evaluation function.
64-bit integer encoding is a performance optimization rather than an effectiveness optimization; it does not directly improve the win rate, but is a prerequisite for achieving search depths of 4 or 5.
Increasing the Expectimax search depth is a means of directly affecting the win rate, but the computational cost increases exponentially as depth increases.
All Facts (5)
A developer attempted to build a 2048 game auto-solver using a genetic algorithm combined with Expectimax search; the current implementation can reach the 2048 tile in 30% to 50% of games.
50%UnverifiedExpectimax is more suitable than Minimax for 2048 because new tiles in 2048 are randomly generated and there is no opponent's will; Expectimax takes the maximum value at player decision nodes and the expected value at random generation nodes.
50%UnverifiedThe role of genetic algorithms in 2048 solvers is typically to optimize the weights of each feature in the Expectimax evaluation function.
50%Unverified64-bit integer encoding is a performance optimization rather than an effectiveness optimization; it does not directly improve the win rate, but is a prerequisite for achieving search depths of 4 or 5.
50%UnverifiedIncreasing the Expectimax search depth is a means of directly affecting the win rate, but the computational cost increases exponentially as depth increases.
50%