search

flowersteam / Grounding_LLMs_with_online_RL

We perform functional grounding of LLMs' knowledge in BabyAI-Text
MIT License
215 stars 24 forks source link

About the average success rate image of paper. #8

Closed CRLqinliang closed 1 year ago

CRLqinliang commented 1 year ago

Hi, guys, I got a question again. I observed that from Fig.13 to Fig.16 in the paper, it seems like the beginning points of those curves are not starting at frame zero. Do you miss something about that?

ClementRomac commented 1 year ago

Hi, Our agents gather 40 transitions from 32 environments in parallel. During the first collection (1200 transitions), it is very likely that none of the 32 environment rechead the end of their episode. Hence we cannot compute the average success rate yet (we must wait for at least one episode to finish its episode to get the return).