On 1 Mar 2017, at 04:11, Nick Mathewson nickm@alum.mit.edu wrote:
On Tue, Feb 28, 2017 at 6:42 AM, teor teor2345@gmail.com wrote:
Hi, Tim!
This looks pretty plausible to me. Do you have time to write up a quick&dirty python function that actually performs the smoothing?
Well, it's a function, but it's not pretty, because it needs to know all the consensus bandwidth values to find the percentile to do the smoothing.
If so, I can test it out on the January consensuses and measure the impact on compressed diff sizes.
To generate a list of bandwidths, run: cat cached-microdesc-consensus | grep ^w | cut -d= -f2 | cut -d" " -f1 > bandwidth_list
To generate a list of pairs: bandwidth rounded_bandwidth
Run: ./bwround.py bandwidth_list
There's an alternative scheme available in the script, to use it, comment out the line: rounding_configs = excess_rounding
(The alternative scheme rounds the 99th percentile to 2 significant figures rather than 3.)
I'd be interested in how each scheme performs.
T
-- Tim Wilson-Brown (teor)
teor2345 at gmail dot com PGP C855 6CED 5D90 A0C5 29F6 4D43 450C BA7F 968F 094B ricochet:ekmygaiu4rzgsk6n xmpp: teor at torproject dot org ------------------------------------------------------------------------