有效地在长度为N的列表中生成长度为K的循环移位的所有排列

时间:2015-10-05 16:39:34

标签: python algorithm permutation cyclic

  

如何在长度为n的列表中生成长度为k的循环移位的所有排列。这里的转变是循环的,正确的。请注意:

如果K == 1,则没有转变。因此,没有那些0班次的排列 如果K == 2,这相当于交换元素。所以n!可以生成排列。

例如。如果列表是[1 4 2],则K = 2(因此从0到N-K,循环)

P1: [1,4,2] #Original list. No shift.
P2: [4,1,2] #Shift from 0 of [1,4,2]
P3: [4,2,1] #Shift from 1 of [4,1,2] as 0 gives P1
P4: [2,4,1] #Shift from 0 of [4,2,1]  
P5: [2,1,4] #Shift from 1 of [1,4,2] as 0 of P4=P3
P6: [1,2,4] #Shift from 0 of [2,1,4]

如果K == 3,事情变得有趣,因为遗漏了一些排列。

例如。如果list = [1,3,4,2],K = 3(因此从索引0到4-3,循环)

P1 : [1,3,4,2] #Original list. No shift. 
P2 : [4,1,3,2] #Shift from 0th of [1,3,4,2]  
P3 : [3,4,1,2] #Shift from 0th of [4,1,3,2]  
P4 : [3,2,4,1] #Shift from 1th of [3,4,1,2] as 0th gives P1
P5 : [4,3,2,1] #Shift from 0th of [3,2,4,1] 
P6 : [2,4,3,1] #Shift from 0th of [4,3,2,1] 
P7 : [2,1,4,3] #Shift from 1th of [2,4,3,1] as 0th gives P3
P8 : [4,2,1,3] #Shift from 0th of [2,1,4,3] 
P9 : [1,4,2,3] #Shift from 0th of [4,2,1,3] 
P10: [2,3,1,4] #Shift from 1th of [2,1,4,3] as 0 from P9=P7,1 from P9=P1,1 from P8=P5  
P11: [1,2,3,4] #Shift from 0th of [2,3,1,4] 
P12: [3,1,2,4] #Shift from 0th of [1,2,3,4] 

#Now,all have been generated, as moving further will lead to previously found values.
  

请注意,这些排列是应该(24)的一半()。   为了实现这个算法,我目前正在使用回溯。这是我到目前为止所尝试的(在Python中)

def get_possible_cyclic(P,N,K,stored_perms): #P is the original list
    from collections import deque  

    if P in stored_perms:
        return    #Backtracking to the previous

    stored_perms.append(P)

    for start in xrange(N-K+1):
        """
        Shifts cannot wrap around. Eg. 1,2,3,4 ,K=3
        Recur for  (1,2,3),4 or 1,(2,3,4) where () denotes the cycle
        """
        l0=P[:start]                    #Get all elements that are before cycle ranges
        l1=deque(P[start:K+start])      #Get the elements we want in cycle
        l1.rotate()                     #Form their cycle
        l2=P[K+start:]                  #Get all elements after cycle ranges

        l=l0+list(l1)+l2                #Form the required list
        get_possible_cyclic(l,N,K,stored_perms)

    for index,i in enumerate(stored_perms):    
        print i,index+1

get_possible_cyclic([1,3,4,2],4,3,[])
get_possible_cyclic([1,4,2],3,2,[])
  

这会产生输出

[1, 3, 4, 2] 1
[4, 1, 3, 2] 2
[3, 4, 1, 2] 3
[3, 2, 4, 1] 4
[4, 3, 2, 1] 5
[2, 4, 3, 1] 6
[2, 1, 4, 3] 7
[4, 2, 1, 3] 8
[1, 4, 2, 3] 9
[2, 3, 1, 4] 10
[1, 2, 3, 4] 11
[3, 1 ,2, 4] 12

[1, 4, 2] 1
[4, 1, 2] 2
[4, 2, 1] 3
[2, 4, 1] 4
[2, 1, 4] 5
[1, 2, 4] 6
  

这正是我想要的,但速度要慢得多,因为这里递归深度超过N> 7。我希望,我已经清楚地解释了自己。有没有优化的人?

1 个答案:

答案 0 :(得分:1)

支票

if P in stored_perms:
随着stored_perms的增长,

变得越来越慢,因为它需要一次比较Pstored_perms的元素,直到找到副本或列表的结尾为止遇到了。由于每个排列都会被添加到stored_perms一次,因此与P进行比较的次数至少是找到的排列数的二次方,这通常是所有可能的排列或其中的一半,具体取决于关于k是偶数还是奇数(假设1

使用set效率更高。 Python的集合基于散列表,因此成员资格检查通常是O(1)而不是O(N)。但是,有一些限制:

  1. 添加到集合中的元素需要为hashable,并且Python列表不可清除。幸运的是,元组是可以清除的,所以一个小小的改动可以解决这个问题。

  2. 对一个集合进行迭代是不可预测的。特别是,在迭代它时,您无法可靠地修改集。

  3. 除了将P更改为元组和stored_perms到集合之外,考虑基于工作队列而不是递归搜索的搜索也是值得的。我不知道它是否会更快,但它避免了递归深度的任何问题。

    把所有这些放在一起,我把以下内容放在一起:

    def get_cyclics(p, k):
      found = set()      # set of tuples we have seen so far
      todo = [tuple(p)]  # list of tuples we still need to explore
      n = len(p)
      while todo:
        x = todo.pop()
        for i in range(n - k + 1):
          perm = ( x[:i]                    # Prefix
                 + x[i+1:i+k] + x[i:i+1]    # Rotated middle
                 + x[i+k:]                  # Suffix
                 )
          if perm not in found:
            found.add(perm)
            todo.append(perm)
      for x in found:
        print(x)