Comment by drdrek | Hacker News Reader

Comment by drdrek | Hacker News Reader

When I want to be random enough I pick a large number than mod by another random number that is bigger then the max option. (mod again in the max number if needed) This seems unpredictable enough I've been unable to see any patterns in it.

lets say you need to pick a number from 1 to 20, if you take 1235 and mod 22. mod is easy to do in your head as you can just subtract in steps until you get there. so 50 times 22 is 1100, we are left with 135. 5 * 22 is 110 so we are left with 25. 25 - 22 is 3.

I picked 3 and I would be hard pressed to guess it would end up 3.

In your example, 1 and 2 are twice as likely as any other number, because of the 22 possible results of n%22, 3 through 20 all only have one result that yields them, but 1 gets generated by 1 and 21, and 2 gets generated by 2 and 22. You could adapt your algorithm by adding "if the result is greater than the range of values you're picking from (i.e. 21 or 22 in your case), try again with a new number."

    >>> import collections, random, pprint
    >>> pprint.pprint(
      sorted(
        list(
          collections.Counter(
            [
              ((random.randint(1,10000000) % 22) % 20)
              for x in range(10000000)
            ]
          ).items()
        )
      )
    )
    [(0, 908920),
    (1, 908264),
    (2, 454167),
    (3, 456019),
    (4, 454551),
    (5, 455183),
    (6, 454127),
    (7, 454308),
    (8, 454939),
    (9, 454602),
    (10, 454963),
    (11, 453117),
    (12, 454046),
    (13, 453812),
    (14, 456243),
    (15, 455025),
    (16, 454101),
    (17, 455072),
    (18, 454409),
    (19, 454132)]

Nice catch, I adopted your sugggested improvement: )

Great method! It's better though to choose prime numbers for mod X operation, that'll make biases less likely.

Reply on news.ycombinator.com