Random Integers from 0 to 999

[Michael Wojcik]

Re: Random Integers from 0 to 999


In article <H_idnWvuAfFCe8 jfRVn-vQ@comcast.com> , Eric Sosman <esosman@acm-dot-org.invalid> writes:[color=blue]
> Michael Wojcik wrote:[color=green]
> >
> > At any rate, I agree with the basic point: if you're dissatisfied, or
> > think you might be, with rand(), better generators are not hard to
> > come by.[/color]
>
> Since the quality of rand() is implementation-specific,
> "better" is hard to justify in a universal sense.[/color]

Sure. In this case, though, a "better" one is merely one that
produces less dissatisfaction in the subject (since dissatisfaction
with rand() is a precondition). Especially perverse cases aside that
looks eminently achievable, as you yourself suggest:
[color=blue]
> It seems that everybody[*] eventually latches onto a favorite
> substitute for rand().[/color]

There you are, then. Everyone will be able to find a "better"
substitute for rand, on the grounds that they will feel that it's
better, and so it will displease them less. What's programming for,
if not to console programmers?
[color=blue]
> My own not-so-humble opinion is that the C Standard should leave the
> requirements on rand() as weak as they are today, but should describe
> it as a "coarse" source of variates, suitable for "casual" use but
> not for serious work.[/color]

Agreed.

--
Michael Wojcik michael.wojcik@ microfocus.com

What is it with this warm, quiet, nauseating bond between them?
-- Rumiko Takahashi, _Maison Ikkoku_, trans. Mari Morimoto, adapt. Gerard
Jones

[Michael Wojcik]

Re: Random Integers from 0 to 999


In article <kfnwtrc1dkv.fs f@alumnus.calte ch.edu>, Tim Rentsch <txr@alumnus.ca ltech.edu> writes:[color=blue]
> mwojcik@newsguy .com (Michael Wojcik) writes:
>[color=green]
> > Alternatively, George Marsaglia has posted code here for a couple of
> > C implementations of Complementary-Multiply-With-Carry (CMWC) PRNGs,
> > which though not portable (IIRC) should be easy enough to adapt to
> > most C implementations . [snip][/color]
>
> The CMWC code that I was able to find (some of it in CLC postings)
> usually used 'long long'.[/color]

Yes, I think the version I have does too, though if memory serves
there were no great differences in adapting it to use a pair of
unsigned longs, if long has fewer than 64 bits. It's been a while
since I looked at it, though.
[color=blue]
> By the way, I had trouble tracking down web articles on CMWC, because
> most of the web writing on CMWC says complImentary rather than
> complEmentary.[/color]

Oh, well. (Perhaps it *is* "complimentary" : the multiply says
something favorable about its operands, or happens for free. I
suspect this is just a case of homophone confusion, though.)
[color=blue][color=green]
> > On a related note: there was a thread on sci.crypt not long ago
> > regarding producing an unbiased subrange of the PRNG's range (a
> > common task which we were recently discussing here) while consuming
> > the minimum possible number of bits from the generator's output.[/color]
>
> For practical purposes I'm inclined to think it's not worth the
> trouble.[/color]

Probably not, though it's a fun discussion. I can see some areas
where it might be useful, though, such as providing a deterministic-
time unbiased generator for a real-time application or making it
easier to blind the time and power consumption of the PRNG for crypto
purposes.

--
Michael Wojcik michael.wojcik@ microfocus.com

Duck: No secret what's worth a hoot ought to be kept quiet.
Pogo: Secrets is usually perty doggone fascinatin'.
Duck: Egg-zackly ... it's completely illogical to keep a secret secret.
Pogo: An' unfair. -- Walt Kelly

[Richard Bos]

Re: Random Integers from 0 to 999

websnarf@gmail. com wrote:
[color=blue]
> 1. The calls to rand() are not independent
>
> or
>
> 2. With a high degree of confidence you can say there is an
> eccentricity in the distribution of the output probabilities that
> underrepresents zero as an output.
>
> This actually matters to some people, and its an interesting echo to
> the "Can you use C for mathematical purposes?" thread from earlier that
> nobody here seems to be sensitive to the above in any way.[/color]

I dare say that people who use C for mathematical purposes _are_ aware
of the above, and moreover are aware that

Anyone who considers arithmetical methods of producing
random digits is, of course, in a state of sin.
Alfred E. Neumann, 1951

Richard

[Grumble]

Re: Random Integers from 0 to 999

Keith Thompson wrote:
[color=blue]
> If an implementation' s rand() function never returns 0, the implementation
> is either non-conforming or of poor quality.
>
> Do you know of an implementation in which rand() never returns 0?[/color]

I was under the impression that the PRNG discussed in [1] had been
used often in the old days?

[1] "Random number generators: good ones are hard to find"

http://www-scf.usc.edu/~csci105/links/p1192-park.pdf

u_0 = 1
u_{n+1} = 16807*u_n mod (2^31-1)

Come to think of it, an implementor could very well return u_n - 1
and set RAND_MAX to 2^31-3...

--
Regards, Grumble

[Keith Thompson]

Re: Random Integers from 0 to 999

rlb@hoekstra-uitgeverij.nl (Richard Bos) writes:
[snip][color=blue]
> I dare say that people who use C for mathematical purposes _are_ aware
> of the above, and moreover are aware that
>
> Anyone who considers arithmetical methods of producing
> random digits is, of course, in a state of sin.
> Alfred E. Neumann, 1951[/color]

I dare say most of them are aware that the quotation is by John von
Neumann, not the guy on the cover of Mad Magazine.

(There is, of course, a connection between Mad Magazine and computer
science; google Potrzebie Knuth for details.)

--
Keith Thompson (The_Other_Keit h) kst-u@mib.org <http://www.ghoti.net/~kst>
San Diego Supercomputer Center <*> <http://users.sdsc.edu/~kst>
We must do something. This is something. Therefore, we must do this.

[Richard Bos]

Re: Random Integers from 0 to 999

Keith Thompson <kst-u@mib.org> wrote:
[color=blue]
> rlb@hoekstra-uitgeverij.nl (Richard Bos) writes:
> [snip][color=green]
> > I dare say that people who use C for mathematical purposes _are_ aware
> > of the above, and moreover are aware that
> >
> > Anyone who considers arithmetical methods of producing
> > random digits is, of course, in a state of sin.
> > Alfred E. Neumann, 1951[/color]
>
> I dare say most of them are aware that the quotation is by John von
> Neumann, not the guy on the cover of Mad Magazine.[/color]

<g> What, me worry?
[color=blue]
> (There is, of course, a connection between Mad Magazine and computer
> science; google Potrzebie Knuth for details.)[/color]

Or buy Van der Linden's Expert C Programming.

Richard

[Grumble]

Re: Random Integers from 0 to 999

Grumble wrote:
[color=blue]
> Keith Thompson wrote:
>[color=green]
>> If an implementation' s rand() function never returns 0, the
>> implementation is either non-conforming or of poor quality.
>>
>> Do you know of an implementation in which rand() never returns 0?[/color]
>
> I was under the impression that the PRNG discussed in [1] had been
> used often in the old days?
>
> [1] "Random number generators: good ones are hard to find"
> http://www-scf.usc.edu/~csci105/links/p1192-park.pdf
>
> u_0 = 1
> u_{n+1} = 16807*u_n mod (2^31-1)
>
> Come to think of it, an implementor could very well return u_n - 1
> and set RAND_MAX to 2^31-3...[/color]

Has this implementation of rand() been used often?