是的,假设rand()
是均匀分布。我们将通过证明每个输入可以以相等的概率生成每个排列来证明这一点。
N=2 可以很容易地证明。我们将把它画成一棵树,其中孩子代表每个字符串,您可以通过将逗号后面的字符插入最左边的字符串来获得。
0,1 //input where 0,1 represent indices
01 10 //output. Represents permutations of 01. It is clear that each one has equal probability
对于 N,我们将有 N-1 的所有排列,并将最后一个字符随机交换为 N
(N-1 0th permutation),N ..... (N-1 Ith permutation),N ________________________
/ \ / \ \
0th permutation of N 1st permutation.... (I*N)th permutation ((I*N)+1)th permutation .... (I*N)+(I-1)th permutation
这种糟糕的感应应该会导致它分布均匀。
例子:
N=2:
0,1
01 10 // these are the permutations. Each one has equal probability
N=3:
0,1|2 // the | is used to separate characters that we will insert later
01,2 10,2 // 01, 10 are permutations from N-1, 2 is the new value
210 021 012 201 120 102 // these are the permutations, still equal probability
N=4:(弯曲以帮助阅读)
0,1|23
01,2|3 10,2|3
012,3 021,3 210,3 102,3 120,3 201,3
0123 0132 0321 3230 2013 2031 2310 3012
0213 0231 0312 3210 1203 1230 1302 3201
2103 2130 2301 3102 1023 1032 1320 3021
ETC