Good news for PPC Random-Number Generator - Printable Version +- HP Forums (https://www.hpmuseum.org/forum) +-- Forum: HP Calculators (and very old HP Computers) (/forum-3.html) +--- Forum: General Forum (/forum-4.html) +--- Thread: Good news for PPC Random-Number Generator (/thread-16963.html) Pages: 1 2 |
Good news for PPC Random-Number Generator - Namir - 05-17-2021 07:51 PM Hi All, I was reading in the PPC ROM manual and found the label RN that generates random numbers. The PPC ROM manual documents the algorithm used to generate random numbers. I happen to find a random-number testing function in Matalb. I wrote a Matlab function that implements the PPC RN routine and ran a test by generating 1 million random numbers. The PPC RN passed the randomness test with flying colors!! :-) Namir RE: Good news for PPC Random-Number Generator - Gene - 05-17-2021 09:01 PM New Seed = Fractional part ( Previous Seed x 9821 + 0.211327 ) Where seed is a decimal number. This is also the random number generator we put into the 67 FUN rom. Gene RE: Good news for PPC Random-Number Generator - Namir - 05-17-2021 09:19 PM (05-17-2021 09:01 PM)Gene Wrote: New Seed = Fractional part ( Previous Seed x 9821 + 0.211327 ) Yep, that's the one!! Namir RE: Good news for PPC Random-Number Generator - Dave Britten - 05-17-2021 10:56 PM Out of curiosity, how does the good old 997 PRNG hold up in the same Matlab test? RE: Good news for PPC Random-Number Generator - Sylvain Cote - 05-17-2021 11:03 PM (05-17-2021 09:01 PM)Gene Wrote: New Seed = Fractional part ( Previous Seed x 9821 + 0.211327 )Same number generator used in the HP-41 GAMES PAC (LBL "RNDM") RE: Good news for PPC Random-Number Generator - Craig Bladow - 05-17-2021 11:03 PM Namir, Matlab uses double precision floating point by default, does the algorithm still pass if limited to single precision or HP-41 precision? RE: Good news for PPC Random-Number Generator - Albert Chan - 05-17-2021 11:35 PM (05-17-2021 11:03 PM)Craig Bladow Wrote: Matlab uses double precision floating point by default, does the algorithm still pass if limited to single precision or HP-41 precision? Algorithm is really LCG, disguised as floating points. Example, we can rewrite it this way (initial seed in units of millionth) Code: function make_rnd(seed) lua> rnd = make_rnd(123456) -- seed 0.123456 lua> rnd(), rnd(), rnd(), rnd(), rnd() 0.672703 0.82749 0.990617 0.060884 0.153091 Seed calculations are integer based, without rounding errors. RE: Good news for PPC Random-Number Generator - Valentin Albillo - 05-17-2021 11:50 PM . That simple algorithm produces exactly one million pseudo-random integers from 0 to 999999 (or from 0.000000 to 0.999999). After that, it simply repeats the same sequence, which is fine for games and such. I used it to generate 1,000,000 user-selectable random palettes for my Fractval program. On the other hand, the HP-71B RND function provides a trillion rng's before repeating and passes the Spectral test, which is adequate for serious purposes. Just a one-million period is too few for that, the longer the period the better. V. RE: Good news for PPC Random-Number Generator - Namir - 05-18-2021 03:13 AM (05-17-2021 10:56 PM)Dave Britten Wrote: Out of curiosity, how does the good old 997 PRNG hold up in the same Matlab test? It passes the test for randomness (based on 1 million random numbers)! Namir RE: Good news for PPC Random-Number Generator - Namir - 05-18-2021 03:15 AM (05-17-2021 11:03 PM)Craig Bladow Wrote: Namir, I don't know. But let me ask which HP-41C application will need to generate more than a million random number? Namir RE: Good news for PPC Random-Number Generator - Massimo Gnerucci - 05-18-2021 06:02 AM (05-17-2021 11:03 PM)Sylvain Cote Wrote:(05-17-2021 09:01 PM)Gene Wrote: New Seed = Fractional part ( Previous Seed x 9821 + 0.211327 )Same number generator used in the HP-41 GAMES PAC (LBL "RNDM") Yep! That's an old friend. RE: Good news for PPC Random-Number Generator - Namir - 05-18-2021 10:35 AM I also tested the following algorithms that were mentioned in HP documentations: f = frac((pi+ r)^3) ---> failed randomness test f = frac((pi+ r)^5) ---> passed randomness test Namir RE: Good news for PPC Random-Number Generator - Craig Bladow - 05-19-2021 01:48 AM (05-18-2021 03:15 AM)Namir Wrote:(05-17-2021 11:03 PM)Craig Bladow Wrote: Namir, The application that tests if there are 1,000,000 unique numbers from this algorithm? RE: Good news for PPC Random-Number Generator - Valentin Albillo - 05-19-2021 02:18 AM (05-18-2021 03:15 AM)Namir Wrote: But let me ask which HP-41C application will need to generate more than a million random number? This 98-step RPN program generates and uses half a million random numbers in the included worked example and would obtain greater accuracy if using several million, matter of fact I ran it using 100 million random values. It wouldn't obtain such accuracy if using a generator with shorter period, the results would be unacceptably biased and thus useless. V. RE: Good news for PPC Random-Number Generator - Namir - 05-19-2021 03:29 AM (05-19-2021 01:48 AM)Craig Bladow Wrote:(05-18-2021 03:15 AM)Namir Wrote: I don't know. But let me ask which HP-41C application will need to generate more than a million random number? In Matlab I sorted the million random numberand checked each number with the next 1000 numbers. There were no matches! RE: Good news for PPC Random-Number Generator - Craig Bladow - 05-19-2021 01:20 PM (05-19-2021 03:29 AM)Namir Wrote:(05-19-2021 01:48 AM)Craig Bladow Wrote: The application that tests if there are 1,000,000 unique numbers from this algorithm? I ran it on a DM41X in FAST mode, plugged into USB power, and after many hours it let me know there was a repeat of the initial seed after 1,000,000 unique numbers. RE: Good news for PPC Random-Number Generator - Thomas Okken - 05-19-2021 03:18 PM (05-17-2021 09:01 PM)Gene Wrote: New Seed = Fractional part ( Previous Seed x 9821 + 0.211327 ) Not to mention the HP-41C Standard Applications book. I'd think this would be the best-known PRNG among HP-41 veterans. Page 24: Quote:Another interesting portion of this program is the random number generator: RE: Good news for PPC Random-Number Generator - Valentin Albillo - 05-19-2021 03:23 PM (05-19-2021 01:20 PM)Craig Bladow Wrote:(05-19-2021 03:29 AM)Namir Wrote: In Matlab I sorted the million random numberand checked each number with the next 1000 numbers. There were no matches! Try it in Free42, it will run orders of magnitude faster. Free42, though a simulator with far greater accuracy, exactly mimics the HP-71B 12-digit RND results, I tried both RAN vs RND sequences for 100 mllion random numbers and they matched perfectly. V. RE: Good news for PPC Random-Number Generator - rprosperi - 05-19-2021 05:41 PM On these extended test runs to verify generated numbers are not repeated, where are all the generated numbers stored, while verifying subsequent numbers don't match, or are you trusting some algorithm's verification? Not prodding here, I really don't know. RE: Good news for PPC Random-Number Generator - Valentin Albillo - 05-19-2021 06:31 PM (05-19-2021 05:41 PM)rprosperi Wrote: On these extended test runs to verify generated numbers are not repeated, where are all the generated numbers stored, while verifying subsequent numbers don't match, or are you trusting some algorithm's verification For this particular algorithm and others like it, if the initial seed ever gets repeated then the whole sequence repeats again so a simple verification requires just to compare each generated number to the initial seed while keeping count of how many numbers are generated. Here, any seed you care to use gets repeated after exactly one millon generated numbers. No need to store anything but the initial seed. To make sure there are no seeds which result in shorter periods, there are more ellaborated yet still simple algorithms that also don't need storing the whole sequence, see this link, "cycle detection". Best regards. V. |