Work-in-progress: Corrected Self Imitation Learning via Demonstrations