Understanding the Score
The score in 2048 is the cumulative sum of all tile merges throughout the entire game — not just the value of the maximum tile at the end.
For example: When two 2-tiles merge into a 4, you earn 4 points. When two 4-tiles merge into an 8, you earn 8 points. The final score represents the total of all these merges from start to finish.
Algorithm Explanations
The following algorithms were tested to determine optimal strategies for playing 2048 across different grid sizes.
Monte Carlo
A probabilistic algorithm that uses random sampling to evaluate possible moves. It simulates many random games from the current state and chooses the move that leads to the best average outcome across all simulations.
Expectimax Tree
A game tree search algorithm designed for games with random elements. Unlike Minimax, Expectimax considers chance nodes (random tile spawns) and computes expected values rather than worst-case scenarios.
IDDFS Graph Search
Iterative Deepening Depth-First Search with graph-based state management. Combines the space efficiency of DFS with the completeness of BFS, incrementally exploring deeper levels while avoiding revisiting duplicate states.
Reinforcement Learning
A machine learning approach where an agent learns optimal strategies through trial and error. The algorithm develops a policy by receiving rewards for successful moves and adjusting its behavior over many training episodes.
Grid Sizes Tested
Standard grid (16 cells)
Grid Area: 16Extended grid (25 cells)
Grid Area: 25Research Context
This dashboard presents data from systematic testing of AI algorithms playing 2048. Each algorithm was run multiple times on different grid configurations to evaluate:
- Average score achieved per algorithm
- Consistency of performance across runs
- How grid size affects algorithm effectiveness
- Whether any algorithm dominates across all conditions
Explore the Dashboard for interactive analysis or visit Research for academic references.