Question buffer_size in yaml & stacked vectors in Behavior Parameters

Hello, I am a student studying multi-agent.
I am looking at the document of ml-agent and I do not understand it, so I ask a question.



In the document, buffer_size is about how many experiences. And as far as I know, the stacked vector of Unity behavior parameters determines whether the agent has an experience (if the stacked vector is 1, experience is not referenced).

So I wonder whether buffer_size in yaml is meaningful or not, when stacked vector is 1?

Please let me know if I am misunderstanding something. Thank you so much.

 

AFAIK the stacked vector only contains observations from previous time steps and is fed into the current step along with current observations. When stacked vector is 1 the agent only receives observations for the current time step. The experiences stored in the buffer may contain these observations but also contain the actions and rewards which are not contained in the vector stack.

So yes buffer size is (extremely) meaningful to learning irrespective of vector stack size.