tutorials-qtictactoe-2.py

import random

from q import Q
from tictactoe import TicTacToe


class Agent:
    def __init__(self):
        self.eps = 1.0
        self.qlearner = Q()

    def _get_action(self, state, valid_actions):
        if random.random() < self.eps:
            return random.choice(valid_actions)
        best = self.qlearner.get_best_action(state)
        if best is None:
            return random.choice(valid_actions)
        return best

    def _learn_one_game(self):
        game = TicTacToe(render=False)
        while True:
            state = game.get_state()
            action = self._get_action(state, game.get_valid_actions())
            winner = game.play(*action)

            if winner or game.is_ended():
                self.qlearner.update(state, action, game.get_state(), 100)
                break

            winner = game.play(*random.choice(game.get_valid_actions()))
            if winner or game.is_ended():
                self.qlearner.update(state, action, game.get_state(), -100)
                break
            self.qlearner.update(state, action, game.get_state(), 0)

    def learn(self, n=20000):
        for _ in range(n):
            self._learn_one_game()
            self.eps -= 0.0001