exp4p.py 文件源码

def _exp4p_score(self, context):
        """The main part of Exp4.P.
        """
        advisor_ids = list(six.viewkeys(context))

        w = self._modelstorage.get_model()['w']
        if len(w) == 0:
            for i in advisor_ids:
                w[i] = 1
        w_sum = sum(six.viewvalues(w))

        action_probs_list = []
        for action_id in self.action_ids:
            weighted_exp = [w[advisor_id] * context[advisor_id][action_id]
                            for advisor_id in advisor_ids]
            prob_vector = np.sum(weighted_exp) / w_sum
            action_probs_list.append((1 - self.n_actions * self.p_min)
                                     * prob_vector
                                     + self.p_min)
        action_probs_list = np.asarray(action_probs_list)
        action_probs_list /= action_probs_list.sum()

        estimated_reward = {}
        uncertainty = {}
        score = {}
        for action_id, action_prob in zip(self.action_ids, action_probs_list):
            estimated_reward[action_id] = action_prob
            uncertainty[action_id] = 0
            score[action_id] = action_prob
        self._modelstorage.save_model(
            {'action_probs': estimated_reward, 'w': w})

        return estimated_reward, uncertainty, score