Learning Chess from Data

Explore how machine learning and Hadoop can be used to rediscover chess rules and gain new insights into the game in this 22-minute EuroPython Conference talk. Delve into the fascinating question of whether watching chess games is sufficient to deduce the rules. Examine the common denominators between different plays and game endings. Investigate the potential of empirical samples to reveal the bigger picture and determine if chess game descriptions provide enough data to understand rules such as legal piece moves, castling, and the difference between check and checkmate. Analyze which features are crucial in describing a chess game and what constitutes a good representation for learning purposes. Consider the minimal sample size required for effective learning and potential pitfalls. Extend the discussion to broader applications, questioning whether big systems can be understood based solely on empirical samples and if physics can be reverse-engineered without external knowledge. Cover topics including introduction, data analysis, simple moves, board status, tradeoffs, checkmate, and different versions of the game.

Introduction
What will we know
The data
Simple moves
Board status
Tradeoff
Checkmate
Versions