Forecasting MCTS Variant Outcomes Across Board Games
MCTS is a widely used search algorithm for developing agents that can play board games intelligently. Over the past two decades, researchers have proposed dozens, if not hundreds, of MCTS variants. Despite this, it's been challenging to determine which variants are best suited for specific types of games. The objective was to create a model to predict how well one Monte-Carlo tree search (MCTS) variant will do against another in a given board game, based on a list of features describing the game. This challenge figured out which MCTS variants work best in different types of games, so we can make more informed choices when applying these algorithms to new problems. Submissions were evaluated using Root Mean-squared error (RMSE) metric.
- 📁 Kaggle Competition: UM - Game-Playing Strength of MCTS Variants