evaluator.py 文件源码

def evaluate_and_update_max_score(self, t, episodes):
        eval_stats = eval_performance(
            self.env, self.agent, self.n_runs,
            max_episode_len=self.max_episode_len, explorer=self.explorer,
            logger=self.logger)
        elapsed = time.time() - self.start_time
        custom_values = tuple(tup[1] for tup in self.agent.get_statistics())
        mean = eval_stats['mean']
        values = (t,
                  episodes,
                  elapsed,
                  mean,
                  eval_stats['median'],
                  eval_stats['stdev'],
                  eval_stats['max'],
                  eval_stats['min']) + custom_values
        record_stats(self.outdir, values)
        if mean > self.max_score:
            update_best_model(self.agent, self.outdir, t, self.max_score, mean,
                              logger=self.logger)
            self.max_score = mean
        return mean