Learning Machine Learning with a Game

AIs playing strategic games have always fascinated humans. Specifically, the reinforcement learning technique Alpha Zero (D.Silver, 2016) has gained much attention for its capability to play Go, which was hard crack problem AI a long time. Additionally, we see rise of explainable (xAI), tries address that many modern decision techniques are black-box approaches and incomprehensible Combining board game relatively simple Connect-Four with explanation offers possibility something about an AI's inner workings itself. This paper explains how combine Alpha-Zero-based known used in supervised learning. this visualization trees. Alpha-Zero combines neuronal network Monte-Carlo-Search-Tree. The approach present focuses on two explanations. first is dynamic analysis evolving situation, primarily based tree aspect, works radial representation (Yee et al., 2001). second static identify relevant situation elements using Lime (Local Interpretable Model Agnostic Explanations) (Christoforos Anagnostopoulos, 2020). aspect. straightforward application towards Monte-Carlo-Search-Tree would be too compute-intensive interactive applications. We suggest modification accommodate search trees sacrifice model agnosticism specifically. use weighted Lasso-based different constellations analyzed by get final situation. Finally, visually interpret resulting linear weights from Lasso board. implementation done Python PyGame library interaction implementation. implemented networks PyTorch Scikit Learn. provides details experimental machines learn game.