search

hongzimao / pensieve

Neural Adaptive Video Streaming with Pensieve (SIGCOMM '17)
http://web.mit.edu/pensieve/
MIT License
524 stars 280 forks source link

question about TRAIN_SEQ_LEN and TOTAL_VIDEO_CHUNCK #62

Open qq1297504766 opened 5 years ago

qq1297504766 commented 5 years ago

Dear Hongzi, There are two variables (TRAIN_SEQ_LEN and TOTAL_VIDEO_CHUNCK) in pensieve. In sim/multi_agent.py, TRAIN_SEQ_LEN = 100 and TOTAL_VIDEO_CHUNCK = 48. In multi_video_sim, the number of video chunks for each video was randomly generated from [20, 100] and TRAIN_SEQ_LEN = 200. Could you explain why is the value of TRAIN_SEQ_LEN larger than TOTAL_VIDEO_CHUNCK? If TOTAL_VIDEO_CHUNCK = 500, then the value of TRAIN_SEQ_LEN should be greater than 500?

Thank you

hongzimao commented 5 years ago

I don't think there is a TOTAL_VIDEO_CHUNCK = 48 in multi-sim code. About which value is actually effective, I think you get that the number of video chunks is bounded in [20, 100]. https://github.com/hongzimao/pensieve/blob/master/multi_video_sim/multi_agent.py#L332 200 wouldn't be triggered because it will hit the end of video first.

You can try making the number of chunks larger. One potential issue you might hit is the variance introduced by the video length. We recently show that this realization actually has broader implications: https://arxiv.org/abs/1807.02264