百木园-与人分享,
就是让自己快乐。

排序:用python实现各种数据结构

 

目录

  • 快速排序
  • 选择排序
  • 插入排序
  • 归并排序
  • 堆排序heapq模块
  • 队列
  • 二分查找

1.快速排序

 

    def quick_sort(_list):
            if len(_list) < 2:
                return _list
            pivot_index = 0
            pivot = _list(pivot_index)
            left_list = [i for i in _list[:pivot_index] if i < pivot]
            right_list = [i for i in _list[pivot_index:] if i > pivot]
        return quick_sort(left) + [pivot] + quick_sort(right)


 

2.选择排序

    def select_sort(seq):
        n = len(seq)
        for i in range(n-1)
        min_idx = i
            for j in range(i+1,n):
                if seq[j] < seq[min_inx]:
                    min_idx = j
            if min_idx != i:
                seq[i], seq[min_idx] = seq[min_idx],seq[i]


3.插入排序

 

    def insertion_sort(_list):
        n = len(_list)
        for i in range(1,n):
            value = _list[i]
            pos = i
            while pos > 0 and value < _list[pos - 1]
                _list[pos] = _list[pos - 1]
                pos -= 1
            _list[pos] = value
            print(sql)


4.归并排序

   

 

 def merge_sorted_list(_list1,_list2):   #合并有序列表
        len_a, len_b = len(_list1),len(_list2)
        a = b = 0
        sort = []
        while len_a > a and len_b > b:
            if _list1[a] > _list2[b]:
                sort.append(_list2[b])
                b += 1
            else:
                sort.append(_list1[a])
                a += 1
        if len_a > a:
            sort.append(_list1[a:])
        if len_b > b:
            sort.append(_list2[b:])
        return sort

    def merge_sort(_list):
        if len(list1)<2:
            return list1
        else:
            mid = int(len(list1)/2)
            left = mergesort(list1[:mid])
            right = mergesort(list1[mid:])
            return merge_sorted_list(left,right)


5.6.堆排序heapq模块

 

    from heapq import nsmallest
    def heap_sort(_list):
        return nsmallest(len(_list),_list)


6.栈

    from collections import deque
    class Stack:
        def __init__(self):
            self.s = deque()
        def peek(self):
            p = self.pop()
            self.push(p)
            return p
        def push(self, el):
            self.s.append(el)
        def pop(self):
            return self.pop()


7.队列

 

 

    from collections import deque
    class Queue:
        def __init__(self):
            self.s = deque()
        def push(self, el):
            self.s.append(el)
        def pop(self):
            return self.popleft()


8.二分查找

    def binary_search(_list,num):
        mid = len(_list)//2
        if len(_list) < 1:
            return Flase
        if num > _list[mid]:
            BinarySearch(_list[mid:],num)
        elif num < _list[mid]:
            BinarySearch(_list[:mid],num)
        else:
            return _list.index(num)

 


来源:https://blog.csdn.net/qq_43159930/article/details/123234038
本站部分图文来源于网络,如有侵权请联系删除。

未经允许不得转载:百木园 » 排序:用python实现各种数据结构

相关推荐

  • 暂无文章