Question

试图了解生日悖论。使用以下代码，我发现我需要平均 12个样本才能发生生日碰撞无法理解为什么它比正常 23人低得多，以获得1/2生日碰撞的机会。即使我使用PyCrypto的StrongRandom，结果也不会改变。

from random import randint
from Crypto.Random.random import StrongRandom
EXPERIMENTS_NUM = 10000
SET_SIZE = 365
SUBSET = 23

where_collision_found = list()
rnd = StrongRandom()
for experiment in range(EXPERIMENTS_NUM):
  for i in range(1,SET_SIZE + 2):
    collision_found = False
    #Generate a subset
    # subset = [rnd.randint(1, SET_SIZE) for x in range(i)]
    subset = [randint(1, SET_SIZE) for x in range(i)]
    # Check for collision
    flags = [False for x in range(SET_SIZE + 1)]
    for k in range(i):
      if flags[subset[k]]: #Collision found
        collision_found = True
      else:
        flags[subset[k]] = True

    if collision_found:
      # print 'Collision found in set:', subset
      break
  where_collision_found.append(i)
print 'average collision:', sum(where_collision_found) / float(len(where_collision_found)), 'in', EXPERIMENTS_NUM, 'experiments'

结果：
average collision: 12.1277 in 10000 experiments

Answer 1

我不清楚你的代码在做什么。这就是我刚刚做的事情：

from random import randrange
N = 365
ns = []
for _ in range(10000):
    n = 0
    seen = set()
    while True:
        b = randrange(N)
        n += 1
        if b in seen:
            break
        seen.add(b)
    ns.append(n)
print(sum(ns) / float(len(ns)))

来自样本运行的输出：

24.6577

这很好。您期望的“23”是分布的中位数;平均值（平均值）预计为24.61659 ...见这里： https://en.wikipedia.org/wiki/Birthday_problem#Average_number_of_people

碰撞需要多少样品（生日悖论）

1 个答案: