I'm making a networked game in c++ that has a lot of particles (not just for looks, they are a gameplay thing). These particles use tons of calls to rand() to determine their movement. When a game starts, I've been using srand() to start a new sequence of numbers for everyone with a shared seed. Due to the number of particles in the game, I am using a lock-step system, which also allows me to ensure an identical sequence of calls to rand() for each client. However, I would also like to use random numbers in displaying these particles, which means that I need a second rand(), essentially. Since I know of no way to use rand() in this way, I thought now might be a good time to find a new way to generate random numbers. Since this is a game, I only need a nice pseudo-random number generator. However, I guess what I need is a RNG class or something, so I can use more than one RNG object at once. I just read that Boost has a random number generator class of some kind, so I was thinking of looking into that. What are your suggestions? If I could create my own, without need for adding another library to my project, that would be great. any recommended reading for creating my own pseudo-RNG? I've done some more research, and the Mersenne twister sounds pretty cool. I also read about WELL, which sounds like it might be even better, but I couldn't tell for sure. I would like to keep the possibility of one day selling this game (or some other game I write using some of this code). It looks like WELL is not for commercial use, but the Mersenne twister is. Is Mersenne probably good enough for my purposes? I'm guessing that it probably is. [Edited by - Dark_Oppressor on March 30, 2010 3:11:07 PM]

Boost.random is fine, especially if you're already using Boost in your project. Alternatively, implementing LCG or MT or pretty much any non-crypto-grade PRNG is pretty trivial.

Thanks for the reply. I've got the Mersenne Twister implemented. I am using it to find a random range between 2 numbers. Here is the equation I'm using:



where lownum and highnum are the two numbers taken as parameters, and range is highnum-lownum+1.

I was reading up on this some more, and I've found some people saying that using the modulus like this might not give as good results as something along the lines of



Is there any merit to that?

The twister should be fine for your needs and i'm sure that minor tweak to it gives better number distribution, but that for your needs you won't even notice a difference.

For what you are doing, you don't need perfection, you just need "good enough", so give it a shot and if you have a problem (ie the particles are noticeably clustering in a way that looks bad) go from there (:

Some good reference...
http://www.agner.org/random/

Quote:Original post by Atrix256
For what you are doing, you don't need perfection, you just need "good enough", so give it a shot and if you have a problem (ie the particles are noticeably clustering in a way that looks bad) go from there (:

This. You'll find that PRNG geeks have a much stricter definition of what's "random looking" than you or I. Your bog-standard LCG will show real, clearly visible artifacts under some of the most normal of situations. Pretty much any other generator is very unlikely to.

Quote:Original post by Dark_Oppressor
Thanks for the reply. I've got the Mersenne Twister implemented. I am using it to find a random range between 2 numbers. Here is the equation I'm using:

Don't do that. Instead, use the supplied uniform distribution stuff to generate the range of values you need.

http://www.boost.org/doc/libs/1_42_0/libs/random/index.html (see very first example)

If you create a distribution with exactly the numbers you need, you don't have to worry about the modulus or any of that. Obviously if you find yourself needing a wide variety of different ranges it gets a bit fiddly compared to modulus, but you can wrap it up in a function if you need.

It also paves the way for you to consider the more interesting distributions for other purposes, should you wish.

Quote:Original post by Sneftel

Quote:Original post by Atrix256
For what you are doing, you don't need perfection, you just need "good enough", so give it a shot and if you have a problem (ie the particles are noticeably clustering in a way that looks bad) go from there (:

This. You'll find that PRNG geeks have a much stricter definition of what's "random looking" than you or I. Your bog-standard LCG will show real, clearly visible artifacts under some of the most normal of situations. Pretty much any other generator is very unlikely to.

This reminds me of a blog post... http://motls.blogspot.com/2009/04/perceiving-randomness-egalitarian-bias.html

Quote:Original post by Kylotan
It also paves the way for you to consider the more interesting distributions for other purposes, should you wish.

I always thought it was awesome that you can actually generate common alternative distributions like normal, logarithmic, etc by combining successive calls to rand with various operators like +, -, and *. And the results are often much faster than algorithms that generate these distributions natively.

I don't know how mathematically sound such methods are, but often mathematical soundness is not as important as speed. Gnu plot seems to suggest they're pretty accurate though.