Heuristic Search in Deterministic 2048

Course project for 237-2-5513 Search Methods in Artificial Intelligence, Ben-Gurion University of the Negev.

We study DET2048 — a fully deterministic 2048 variant (every move spawns a 2-tile in the top-right-most empty cell) — as a domain for heuristic search: merge-chain heuristics, milestone-restart IDA* with weighted move costs, and greedy depth-limited lookahead. Full details in the report.

Structure

Path	Contents
report/report.pdf	The lab report (AAAI camera-ready layout)
report/report.tex	LaTeX source
curr_version/2048solver/	Game engine, solvers, Pygame GUI
experiments/	Experiment harness, raw results (CSV), figures

Running the game / solvers

cd curr_version/2048solver
python3 gui.py          # Pygame GUI: pick board size, solver mode, weights

Reproducing the experiments

cd experiments
python3 experiments.py all   # runs all suites; resumable, one CSV row per config
python3 plots.py             # rebuilds all report figures from results/*.csv

The game is deterministic, so every configuration is a single exact run. Each run is capped by a wall-clock budget (35 s in the comparison/depth/target suites, 20 s in weight tuning, 40 s in the final transfer runs); runs exceeding the budget are recorded as timeout.

Compiling the report

cd report
pdflatex report.tex && pdflatex report.tex