Mapping from chunk size to bitrate

[jlabhishek]

Hello, I have a doubt in section 4.3)Enhancement for multiple videos The paper says that "Then, to determine the input state for a specific video, we take the chunk sizes and map them to the index which has the closest bitrate." Can you share some information regarding how it it done, the synthetic dataset created has a mask present in it, but how do we fill in that mask values for real data.

Please let me know if there is some doubt in the question Thank you

[hongzimao]

If I understand what you were asking, the relevant code is in the beginning of multi_agent.py https://github.com/hongzimao/pensieve/blob/master/multi_video_sim/multi_agent.py#L37-L60

[jlabhishek]

Hello hongzi, Actually my questions is related to how the mask is created, let my explain through an example Suppose the training is done with: VIDEO_BIT_RATE = [200,300,450,750,1200,1850,2850,4300,6000,8000] Now I have a new video which has the following bitrates : [1525, 2200,5500] which are not exactly found in the model bitrates, how can we create the mask then? Please let me know your thoughts.

Thank you

[hongzimao]

If I remember, our way of modeling multi-videos doesn't capture this case. I would imagine you do some nearest neighbor mapping and the mask out the rest bitrates. In your example, I would map 1525 to action 5 (VIDEO_BIT_RATE[5] = 1850) and so on. At the end you just want to map the bitrate to a subset of the actions.