Repository for a Maastricht University Master's Project for explaining Monte Carlo and A* tree search for the SameGame game and it's variations
Explaining why intelligent search agents such as Monte Carlo Tree Search (MCTS) chose an action over another is often not easily and intuitively explainable. Moreover, differences in scoring values provided by intelligent search agents do not suffice to explain why an action was chosen over another. Therefore, we present a method of finding relevant features for the given game SameGame played by MCTS and putting these scoring differences into human understandable explanations. Relevancy of a feature was determined by the value of these features compared to other available actions and combining it with offline correlation analysis between the feature values and the scores given by MCTS. Additionally, explaining the moves chosen by MCTS with a decision tree was explored.