generater big rand number

[remlostime]

i use g++ to generater rand number, now i find that the RAND_MAX is
32367 in my computer, how can i make a bigger rand number( the number
is wihin in the integer(2^32-1))

 

[peter koch]

i use g++ to generater rand number, now i find that the RAND_MAX is
32367 in my computer, how can i make a bigger rand number( the number
is wihin in the integer(2^32-1))

You have several options: one is to use a library such as boost which
gives you several options to choose different random number generators
of high quality. Another is to compose your number by calling rand
several times. To illustrate: if your random number only gave values
from 0 to 9, but you needed values from 0 to 99, the solution would my
to use rand()*10 + rand(). I am sure you can extrapolate from here ;-)

/Peter

 

[Juha Nieminen]

i use g++ to generater rand number, now i find that the RAND_MAX is
32367 in my computer, how can i make a bigger rand number( the number
is wihin in the integer(2^32-1))

By using a better random number generator? There are tons of them if
you search the internet.

(Curiously it's actually surprisingly difficult to find a good random
number generator which is portable, written in C++, very easy to use and
doesn't require you to install the entire Boost library to simply use
the RNG. For this reason I made this C++ version of the ISAAC rng:
http://warp.povusers.org/IsaacRand.zip )

 

[gw7rib]

You have several options: one is to use a library such as boost which
gives you several options to choose different random number generators
of high quality. Another is to compose your number by calling rand
several times. To illustrate: if your random number only gave values
from 0 to 9, but you needed values from 0 to 99, the solution would my
to use rand()*10 + rand(). I am sure you can extrapolate from here ;-)

I don't think your second approach will work. For example, suppose
your random number generator gives only ten values and produces the
sequence:

1, 6, 4, 7, 9, 0, 3, 5, 2, 8, 1, 6, etc

then your improved version will only ever give the numbers 16, 64, 47,
79, 90, 3, 35, 52, 28 and 81, instead of the full range from 0 to 99.

 

[peter koch]

I don't think your second approach will work. For example, suppose
your random number generator gives only ten values and produces the
sequence:

1, 6, 4, 7, 9, 0, 3, 5, 2, 8, 1, 6, etc

then your improved version will only ever give the numbers 16, 64, 47,
79, 90, 3, 35, 52, 28 and 81, instead of the full range from 0 to 99.

That would be a short sequence - not anything you'd expect from a
quality random generator.

/Peter

 

[James Kanze]

Well, yes, if the generator produces a sequence of length ten,
then there's not much you can do. But producing random digits
from 0 to 9 doesn't mean producing a sequence of length ten.
Typically a random number generator has a much longer period
than that. If the period is long enough, there's no problem
tiling the values, as the second approach suggests.

Yes and no. The sequence should be significantly longer than
the value one is trying to generate; in other words, the value
actually returned by rand() shouldn't represent the entire
internal state of the machine.

For historical reasons, at in some environments, RAND_MAX is
defined as 32767, even though the actual generator uses 31 or 32
bits internally (and has an actual period of around 2^31). If
this is the case, then the proposed technique is fine.
Similarly, rand() is required to return an int---a number of
quality generators use significantly more state internally. In
such cases, the technique is also valid. If the generator's
period is only RAND_MAX, however, it's likely to be too pretty
bad (but depending on the use, maybe "good enough" anyway).

 

[Juha Nieminen]

Agreed; that one seems to assume that unsigned int has exactly 32 bits, no
more, no less. Take a look at the ind macro in IsaacRand.cc line 21.
That's just one non-portability I found after a quick scan.

When I say "portable" I mean things like "doesn't include <windows.h>"
and the like. I have seen *way* too many RNGs out there which use
whatever non-standard headers and non-standard code.

If you are worried that in some system 'int' might not be 32-bit then
put some assert() somewhere or whatever. The actual RNG code is not
mine. I only wrapped it inside the class and removed everything from the
global namespace (the original C code put something like 100 symbols in
the global namespace). It fulfills my needs for a fast high-quality RNG
in both linux and windows just fine. The same cannot be said from the
majority of RNG libraries I have seen.