evaluator.py 文件源码

def evaluate_and_update_max_score(self, t, episodes, env, agent):
        eval_stats = eval_performance(
            env, 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 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)
        with self._max_score.get_lock():
            if mean > self._max_score.value:
                update_best_model(
                    agent, self.outdir, t, self._max_score.value, mean,
                    logger=self.logger)
                self._max_score.value = mean
        return mean