mcts.py contains our implementation of Monte Carlo tree search. Our first class is RootNode, which is meant to represent the root node of the MCTS tree at the start of a simulation. By definition, the root node does not have a parent. Having a separate class for the root node is not absolutely necessary, but it does keep the code cleaner:
import collectionsimport mathimport numpy as npimport utilsfrom config import MCTSPARAMETERS, GOPARAMETERSclass RootNode(object): def __init__(self): self.parent_node = None self.child_visit_counts = collections.defaultdict(float) self.child_cumulative_rewards = collections.defaultdict(float)
Next, we implement the MCTreeSearchNode class. This class has several attributes, the most important ones ...
