For the purposes of this project, a "player" is self-contained and doesn't use any heuristics. A "strategy" requires a heuristic to be passed in, at which point it becomes a player. So the Expectimax strategy with the Corner Building heuristic becomes the ExpectimaxCornerBuilding player.

Each strategy/player is in its own class.

See the results directory for the performances of each strategy.

These players pick moves without precalculating future outcomes.

This player picks a random valid move.

This player randomly picks between left and up. If both moves aren't allowed, it picks randomly between right and down.

This player tries the same move order each time. If left is allowed, it moves left. Otherwise if up is allowed, it moves up. Otherwise if right is allowed, it moves right. Otherwise it moves down.

This player plays the same moves in order. It starts with up, then plays right, then plays down, then plays left, and repeats.

writeup todo

writeup todo

writeup todo

For each possible move, this solver runs t random games from the resulting board and averages the score from all those games. The move that has the best average score is picked.

The time it takes for this solver is approximately linear in t. The rate of improvement starts to decrease as t gets to around a thousand.

This strategy attempts to merge as many tiles as possible at a certain search depth. It's essentially a recursive brute force. Each move checked runs some amount of trials to simulate the randomly placed tiles.

The d (depth) and t (trials) parameters can be changed, although the time to run grows exponentially in d, so usually d <= 4. The number of trials doesn't seem to have much of an effect beyond around 5 or 6.

writeup todo

writeup todo

writeup todo

writeup todo