My dad taught statistics and always did the thing about the odds of two people in class having the same birthday. He's gone now so I turn to the math majors here. How do we compute how often to expect a duplicate or repeat in an rng for n digits or a range? I'm looking at 5 digits, 0..99999.

A good question is never answered. It is not a bolt to be tightened into place but a seed to be planted and to bear more seed toward the hope of greening the landscape of the idea. John Ciardi

google for 'birthday paradox' It works out that the odds hit 50% quickly, I think 22 people or so, where your naive expectation is that it would take 180.

I worked with a guy who could calculate it in his head.

Stan James
(instanceof Sidekick)
Ranch Hand

Joined: Jan 29, 2003
Posts: 8791

posted

0

While working with this I read the code for our custom RNG and got scared.

I started looking for ways to compare randomness and settled on even distribution. I made a unit test that for a parameter n generates n**2 values between 1..n and asserts that 99% of the possible numbers are generated within 10% of the ideal distribution. Java Random flies through this for n over a few hundred with no problem.

The custom generator didn't trust the seed algorithm for some reason and resets the seed to the output value every time. I guess because the output value is many orders of magnitude smaller than the original seed, the value converges on repeating cycles pretty quickly. For 1..10000 38 possible values had counts within 1 of each other, the other 9962 occurred zero or one times.

Moral: You are almost certainly not smarter than the library, or, in this case, Donald Knuth.

Stan, I saw the first part of your post in the Today's Posts section and thought to myself "Better remind him what Knuth said about random-number generators". But... I see that's covered off already. Good. Carry on then.