a3c tic tac toe value search

import numpy as np

import random

import tensorflow as tf

import json

import time

import copy

class TicTacToePolicy(dc.rl.Policy):

    def create_layers(self, state, **kwargs):

        d1 = Flatten(in_layers=state)

        d2 = Dense(

        return {'action_prob': probs, 'value': value}

def main():

def eval_tic_tac_toe(value_weight, games=10 ** 4, rollouts=10 ** 5):

    """

    Returns the average reward over 1k games after 10k rollouts

    :param value_weight:

    :return:

    """

    env = TicTacToeEnvironment()

    policy = TicTacToePolicy()

  a3c = dc.rl.A3C(env, policy, entropy_weight=0, value_weight=0.25)

    a3c = dc.rl.A3C(env, policy, entropy_weight=0.01, value_weight=value_weight)

    a3c.optimizer = dc.models.tensorgraph.TFWrapper(

        tf.train.AdamOptimizer, learning_rate=0.01)

  a3c.fit(100000)

    a3c.fit(rollouts)

    rewards = []

    for i in range(games):

        env.reset()

        reward = -float('inf')

        while not env._terminated:

    print(env.display())

    print(a3c.predict(env._state))

            action = a3c.select_action(env._state)

    print(action)

    print(env.step(action))

  print(env.display())

            reward = env.step(action)

        rewards.append(reward)

    return np.mean(rewards)

def main():

    scores = {}

    value_weight = 0.05

    while value_weight <= 1.0:

        print(value_weight)

        score = eval_tic_tac_toe(value_weight)

        scores[value_weight] = score

        with open('tictactoe_value_search.json', 'w') as fout:

            fout.write(json.dumps(scores))

        value_weight += 0.05

if __name__ == "__main__":