View Quote

#7330 Up↑ /2 Down↓ [Report] 20140911 18:24 GMT
<fitzgen> jimb: sampling w/ probability = .1 adds about 25% overhead of probability = 1.0
<jorendorff> fitzgen: did you try setting the counter initially to a random value, then decrementing?
<jorendorff> fitzgen: and then when it hits zero, log and reset
<fitzgen> jorendorff: that's next, I'm working out the math on paper now
<jimb> jorendorff: You get a different distribution that way.
<jorendorff> jimb, fitzgen: *not necessarily*  I mean, yes, that's true, but perhaps you actually *prefer* a more uniform (not random!) distribution, and you're only perturbing it enough to get rid of resonance
<jimb> jorendorff: I figured out a way to get exactly the same distribution as generating a random number at each event, but which only requires generating random numbers when you actually take a sample.
* jorendorff is so pleased to be a step ahead of jimb on this one
<jorendorff> that never happens
<jimb> jorendorff: You're a step behind.
<fitzgen> ouch
<fitzgen> https://en.wikipedia.org/wiki/List_of_burn_centers_in_the_United_States
<jorendorff> thank you
Kindly hosted by jX and in no way affiliated with the Mozilla Foundation.