Trainer.train(store_episodes=True)is meant to store episode trajectories in itr_0.pkl,itr_1.pkl, ..., but in the current implementation it does not.
The only side effect of store_episodes=True, is on line 443 of trainer.py:
save_episode = (self.step_episode if self._train_args.store_episodes else None)
self.step_episode is then passed to self._stats and saved correctly.
However, self.step_episode is never set in most algorithms.
Trainer.train(store_episodes=True)is meant to store episode trajectories initr_0.pkl,itr_1.pkl, ..., but in the current implementation it does not.The only side effect of
store_episodes=True, is on line 443 oftrainer.py:save_episode = (self.step_episode if self._train_args.store_episodes else None)self.step_episodeis then passed toself._statsand saved correctly.However,
self.step_episodeis never set in most algorithms.