最后,我的代码花费了太长时间来明确计算每个可能的组合,所以我想出了一种方法,只找到唯一的组合,然后分析计算它们的多重性。
它基于以下思想:调用输入列表A和每个子集中的元素个数k。首先对列表进行排序,并将 k 指针初始化为 A 的前 k 个元素。然后反复尝试将最右边的指针向右移动,直到遇到新值。每次移动另一个指针而不是最右边时,所有指向它右边的指针都设置为它的邻居,例如如果指针 1 移动到索引 6,指针 2 移动到索引 7,依此类推。
任何组合 C 的多重性可以通过乘以二项式系数 (N_i, m_i) 来找到,其中 N_i 和 m_i 分别是元素 i 在 A 和 C 中出现的次数。
下面是一种蛮力方法的实现,以及一种利用唯一性的方法。
此图比较了蛮力计数的运行时间与我的方法。当输入列表有大约 20 个元素时,计数变得不可行。
# -*- coding: utf-8 -*-
from __future__ import division
from itertools import combinations
from collections import Counter
from operator import mul
import numpy as np
from scipy.special import binom
def brute(A, k):
'''This works, but counts every combination.'''
A_sorted = sorted(A)
d = {}
for comb in combinations(A_sorted, k):
try:
d[comb] += 1
except KeyError:
d[comb] = 1
#
return d
def get_unique_unordered_combinations(A, k):
'''Returns all unique unordered subsets with size k of input array.'''
# If we're picking zero elements, we can only do it in one way. Duh.
if k < 0:
raise ValueError("k must be non-negative")
if k == 0 or k > len(A):
yield ()
return # Done. There's only one way to select zero elements :)
# Sorted version of input list
A = np.array(sorted(A))
# Indices of currently selected combination
inds = range(k)
# Pointer to the index we're currently trying to increment
lastptr = len(inds) - 1
# Construct list of indices of next element of A different from current.
# e.g. [1,1,1,2,2,7] -> [3,3,3,5,5,6] (6 falls off list)
skipper = [len(A) for a in A]
prevind = 0
for i in xrange(1, len(A)):
if A[i] != A[prevind]:
for j in xrange(prevind, i):
skipper[j] = i
prevind = i
#
while True:
# Yield current combination from current indices
comb = tuple(A[inds])
yield comb
# Try attempt to change indices, starting with rightmost index
for p in xrange(lastptr, -1 , -1):
nextind = skipper[inds[p]]
#print "Trying to increment index %d to %d" % (inds[p], nextind)
if nextind + (lastptr - p) >= len(A):
continue # No room to move this pointer. Try the next
#print "great success"
for i in xrange(lastptr-p+1):
inds[p+i] = nextind + i
break
else:
# We've exhausted all possibilities, so there are no combs left
return