Re: 32-bit random numbers
- Posted by "Igor Kachan" <kinz at peterlink.ru> Jul 11, 2004
- 542 views
Hi Al, ---------- > From: Al Getz <guest at RapidEuphoria.com> > To: EUforum at topica.com > Subject: Re: 32-bit random numbers > Sent: 10 jul 2004 y. 23:36 > > > posted by: Al Getz <Xaxo at aol.com> > > Igor Kachan wrote: > > > > I can say nothing about the distribution > > on your formula. I just see it is not flat. > > Ok, it's one thing to *SAY* it's not flat, > for example, i can *SAY* YOURS isnt flat, > but it's another story to *PROVE* it's not flat. > > You see now? Yes, I see. It is not flat, your distribution. This my affirmation is very common. I can say nothing about concrete parameters of your distribution, it is the same thing that I can say nothing about the distribution on your formula, yes. Formula is yours, try to prove it is good. This is not my problem, Al. I am trying to explain that N must be <<< K in proposed the rand_atom() function. Did you understand my explanation? > You can teach a 4 year old to say > 'everything is not flat' but it's not true, > is it? He he, Al, even an any 40 years old can not understand the Einstein's theory about all curved spaces > In other words, PROVE it's not flat. Al, distribution is yours, we need the flat one, prove please yours one is flat. BTW, it is the very complicated task to create a good flat RNG, very, ask Rob. rand() is one of the good examples. We use it for our tasks. > The reason i say this is because it looks like the > distribution *IS* flat, but there is of course always > the chance i made a mistake in the analysis, im just > human! I'm just human too, it seems to be, no? > So, if you want me to understand why you say > it is *NOT* flat, please show me how you came > to this conclusion. Al, it is very long another [OT] story, I worked on Russian navy research many years, models-shmodels, we used Monte Carlo method, these RNGs-sh-sh-mrngs ... brrrr ... Another story, Al, books-sh-m-books ... Ricardo knows, he says too - not flat. Believe him. > BTW, i wasnt saying YOUR function isnt flat, > im just asking why you needed another function, > that's all, and what the benefits are. We just need a flat discrete distribution in maximum integer range with step = 1. My function seems to give just this one. No another needs. > I thought maybe you wanted higher and higher > random integers so i proposed another solution > The proposed solution goes up to ANY desired integer > ceiling, not just #FFFFFFFF or whatever. > Im not saying there is no chance i made a mistake, > im just asking that you show me why you say > it's not flat, especially the > > N*(rand(N)-1)+rand(N)-1, with N=#10000 > > formula? Al, this formula gives the composition of TWO random numbers, each one with SAME flat distribution. N is not random, *each* call of rand(N) dives unpredictable number from the [1, N] range. Read please some good book on this subject to be sure. It is too long story for this list and especially for my time. > BTW, by 'flat' i assumed you meant that there > is equal chance of getting any number from 0 to max, > so there are no preferences. Yes, it is flat as I understand it. (Good question for that late stage of discussion.) > BTW(2), it would also be good to have a formula > to specify the max, such as max=N*(N-1)+N-1 > or something like that. Maybe, I don't know. > > But you do see just now that proposed new > > function seems to give the flat distribution. > > Yes, but i was only asking what the reasoning behind > developing a new formula would be, if it does the same. No, not the same. N and K. N << K. Then it seems to be flat. It requires the careful testing after discussion. > If it does something different, that's cool too Al, cool or not very cool it is not our question. We need the flat distribution on full EU integer atoms range with step = 1, for now. > Take care, > Al Regards, Igor Kachan kinz at peterlink.ru