hongzimao / pensieve

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

Some discrepancies #71

Closed ghost closed 5 years ago

ghost commented 5 years ago

I found a discrepancy between the paper and the code. your paper says the input of network is past chunk throughput (state[2,:],line 281 in /sim/multi_agent.py) past chunk dowloadtime (state[3,:],line 282 in /sim/multi_agent.py) next chunk sizes (state[4,:],line 283 in /sim/multi_agent.py) current buffer size (state[1,-1],line 280 in /sim/multi_agent.py) number of chunk left (state[5,-1],line 284 in /sim/multi_agent.py) last chunk bitrate (state[0,-1],line 279 in /sim/multi_agent.py)

but,in your define of network(line 65-70 and 163-168 in /sim/a3c.py), input to the network does not contain the number of chunk left. in line 70 and 168,the input is inputs[:,4:5,-1], but inputs[:,4:5,:] is next chunk size,not number of chunk left Is there any particular reason for this?