Top-Ranked Abalone AI Bot
Budget: $30 – $60 USD
The task is to engineer a competition-grade Abalone agent in Python that can climb into the top four of the CodinGame leaderboard. Two distinct engines are required:
• a Minimax/Alpha-Beta searcher featuring modern refinements—iterative deepening, transposition tables, killer moves, Zobrist hashing, and a high-performance evaluation built around board-control heuristics;
• a separately tunable Monte Carlo Tree Search implementation that explores state-of-the-art rollout policies and selection formulas (this second engine does not itself have to reach the leaderboard).
Accuracy and eventual win-rate matter more than raw speed, so every design decision should favour smarter move choice and stable play strength. At minimum the evaluator must quantify board-control advantage; feel free to enrich it with additional signals (marble differential, push potential, centricity, etc.) as long as you justify them in the accompanying write-up.
Deliverables
1. Well-structured Python 3 source code ready to submit on CodinGame.
2. A technical report that walks through algorithmic design, heuristic reasoning, search enhancements, parameter tuning, and any self-play or regression testing you used to reach the target ranking.
I will validate success by submitting the bot myself and confirming a sustained top-four position over at least 100 ranked games.
• a Minimax/Alpha-Beta searcher featuring modern refinements—iterative deepening, transposition tables, killer moves, Zobrist hashing, and a high-performance evaluation built around board-control heuristics;
• a separately tunable Monte Carlo Tree Search implementation that explores state-of-the-art rollout policies and selection formulas (this second engine does not itself have to reach the leaderboard).
Accuracy and eventual win-rate matter more than raw speed, so every design decision should favour smarter move choice and stable play strength. At minimum the evaluator must quantify board-control advantage; feel free to enrich it with additional signals (marble differential, push potential, centricity, etc.) as long as you justify them in the accompanying write-up.
Deliverables
1. Well-structured Python 3 source code ready to submit on CodinGame.
2. A technical report that walks through algorithmic design, heuristic reasoning, search enhancements, parameter tuning, and any self-play or regression testing you used to reach the target ranking.
I will validate success by submitting the bot myself and confirming a sustained top-four position over at least 100 ranked games.