Quote:Original post by Aryabhatta

Quote:Original post by iMalc
From the looks of the question, the OP just wants to know how some magical function could produce a seemingly random number?

It is typically implemented pretty much as follows.

long holdrand;void srand(long seed){	holdrand = seed;}short rand(){	return (short)((holdrand = holdrand * 214013 + 2531011) >> 16);}

Simple eh!

Not really. I think the right shift by 16 bits causes this to be a bad one!
For instance, off hand I guess that this version of rand returns numbers which will almost surely end with 0x26. Instead of taking a mod you are doing a shift. Not the same thing. Taking the lower 16 bits would probably have been better. Though i admit, your code needs more analysis to claim that this one is not a good one :-)...

I think what IMalc wanted to refer to is a Linear Congruential Generator

You unfortunately don't know what you're talking about. I didn't just pull this out of thin air.
This is a standard implementation of a random number generator, taken from a common runtime library. Search for the above two prime numbers on google and you'll most likely get tons of search results with variants of this prng.
The upper 16 bits are far more random than the lower 16.
I know that this is a Linear congruential generator.

Quote:Original post by iMalc
I didn't just pull this out of thin air.
This is a standard implementation of a random number generator, taken from a common runtime library.

It's a standard RNG used by many standard C library implementations, and it's good enough for simple applications, but it's actually a pretty poor selection of parameters. It fails certain tests of randomness and degenerates into very short sequences given certain seed values.

A better selection might be Lewis, Goodman, and Miller's rand0 (a=16807, c=0, m=2147483647) or possibly the Lehmer generator (a=48271, c=0, m=2147483647). These have been thoroughly analysed and found much better than the rand() function.

You can eliminate the LCG's pairwise sawtooth with a lagged fibonacci seived through a discard block filter (google for James's luxury-level-3 integer adaptation of Luescher's generator) or better yet, Matsumoto and Nishimura's Mersenne Twister. Both lagged fibonacci and the mersenne twister are faster, and produce better "randomness" by most accepted definitions, and have longer periods than the good ol' LCG, but they take considerably more memory.

It is important, however, to realize that the bitshift is a better way to select an integral subrange than is using the % operator, which is what most people do.

--smw

Quote:Original post by Aryabhatta

Quote:Original post by iMalc

Quote:Original post by Aryabhatta
<snip>

Though i admit, your code needs more analysis to claim that this one is not a good one :-)...

I think what IMalc wanted to refer to is a Linear Congruential Generator

You unfortunately don't know what you're talking about. I didn't just pull this out of thin air.
This is a standard implementation of a random number generator, taken from a common runtime library. Search for the above two prime numbers on google and you'll most likely get tons of search results with variants of this prng.
The upper 16 bits are far more random than the lower 16.
I know that this is a Linear congruential generator.

Whoa! Slow down there pal...Sorry If I offended you. I didn't mean to.

I did say your code needs more analysis before anyone can claim it is a bad one! What I stated was my opinion.. Hence the words "I think".

Upon rereading the chapter in Knuth, in LCG's the higher order bits are more random than the lower order bits if the modulus equals the word size, which is true in this case. So taking the higher order 16 bits is right.. but this RNG is pretty bad nonetheless.

I'm sorry, I shouldn't have acted so defensively. I've oft heard that it isn't a terrific prng, but one seldom comes across a case where anything more random is needed (especially for a beginner). For example a particle generator animation would look indistinguishable from one with a better prng. It's only when you come to analyse where every individual droplet landed over a long period of time that you'd notice any difference.