a3C.py 文件源码

def choose_action(self):
        if self.config.mode == 'discrete':
            return tf.multinomial(tf.log(self.a_prob), 1)[0][0]  # ???????tf.log??????action_dim??
        elif self.config.mode == 'continuous':
            # axis = 0?????0??squeeze
            sample_action = self.action_normal_dist.sample(1) * self.config.ACTION_GAP + self.config.ACTION_BOUND[0]
            return tf.clip_by_value(tf.squeeze(sample_action, axis=0),
                                    self.config.ACTION_BOUND[0],
                                    self.config.ACTION_BOUND[1])[0]