On Thu, Feb 07, 2002 at 10:57:52AM +0100, Sebastian Sp?th wrote: > Oskar Sandberg wrote: > > On Thu, Feb 07, 2002 at 12:21:41AM -0500, Tavin Cole wrote: > > It should be noted, again, that the deviation on those results is the > > deviation of the mean value over the number of hops made per request, > > not the deviation of the per hop value. > > Ahh, so htlTimeExpected is per hop, while the deviation is per request? > That is not too obvious, mybe we can rename it to requestTimedeviation > or something else (or make it a true *PER_HOP*deviation)?
No, they are both per hop. What I said was that the values being registered with the diagnositcs are of the mean hops per request (that is obviously the only value we have, we cannot know the individual hop times). The mean of a set of means will still estimate the expected value of the distribution, so that much is correct, but the standard deviation of a set of means will NOT correctly estimate the variance of the distribution due to central convergence, so that number cannot be taken directly. > > Does anybody have any wishes regarding other statistical data we should > > keep? Min and max as well as frequency diagrams should be possible > > (though the latter would take a bit of memory) > > Min, Max, Mean, and std deviations (maybe printing them in addition to > the sum, sum of squares) should be sufficient to measure what we need. > > BTW, how do I get the std. dev out of SUM(X_i), SUM(X_i^2) and n? My > stats knowledge is a bit rusty. > > std dev = SQRT(1/(n-1) * SUM((X_i-X_mean)^2*f_i)) > doesn't work directly without some clever stuff. std dev ~= sqrt(SUM(X_i^2)/n - (SUM(X_i) / n)^2) I know a lot of probability theory, but very little statistics, so I'm just treating mean as expectation and stddev as sqrt(variance). I'm pretty sure that that is ok as long as you have enough observations. -- Oskar Sandberg oskar at freenetproject.org _______________________________________________ Devl mailing list Devl at freenetproject.org http://lists.freenetproject.org/mailman/listinfo/devl
