INNER CODE UNIT · Python

policy_evaluation

rlcode/reinforcement-learning · 1-grid-world/1-policy_iteration.py:18

    def policy_evaluation(self):
        next_value_table = [[0.00] * self.env.width
                                    for _ in range(self.env.height)]

        # Bellman Expectation Equation for the every states
        for state in self.env.get_all_states():
            value = 0.0
            # keep the value function of terminal states as 0
            if state == [2, 2]:
                next_value_table[state[0]][state[1]] = value
                continue

            for action in self.env.possible_actions:
                next_state = self.env.state_after_action(state, action)
                reward = self.env.get_reward(state, action)
                next_value = self.get_value(next_state)
                value += (self.get_policy(state)[action] *
                          (reward + self.discount_factor * next_value))

View source record →