The AlphaGo Zero method
Overview
At a high level, the method consists of three components, all of which will be explained in detail later, so don't worry if something is not completely clear from this section:
- We traverse constantly the game tree, using the Monte-Carlo Tree Search (MCTS) algorithm, the core idea of which is to semi-randomly walk down the game states, expanding them and gathering statistics about the frequency of moves and underlying game outcomes. As the game tree is huge, both in terms of the depth and width, we're not trying to build the full tree, just randomly sampling the most promising paths of it (that's the source of the method's name).
- At every moment, we have a best player, which is the model used to generate the data via the self-play. Initially, this model has random weights, so it makes the moves randomly, like a four-year-old just learning how chess pieces move. However, over time, we'll replace this best player with better variations...