Epsilon to loss function

import random

import tensorflow as tf

import time

import copy

from deepchem.models.tensorgraph.layers import Flatten, Dense, SoftMax, \

  Variable, \

      self._state[0][move[0]][move[1]] = TicTacToeEnvironment.O

  def step(self, action):

    self._state = copy.deepcopy(self._state)

    row = action // 3

    col = action % 3

  def create_tensor(self, **kwargs):

    reward, action, prob, value = [layer.out_tensor for layer in self.in_layers]

    log_prob = tf.log(prob)

    log_prob = tf.log(prob + 0.0001)

    policy_loss = -tf.reduce_sum(

        (reward - value) * tf.reduce_sum(action * log_prob))

    value_loss = tf.reduce_sum(tf.square(reward - value))

    for i in range(self.a3c.max_rollout_length):

      if self.env.terminated:

        break

      state = copy.deepcopy(self.env.state)

      state = self.env.state

      for j in range(len(state)):

        states[j].append(state[j])

      feed_dict = _create_feed_dict(self.features, state)