Heim >Backend-Entwicklung >Python-Tutorial >Python判断列表是否已排序的各种方法及其性能分析

Python判断列表是否已排序的各种方法及其性能分析

WBOY
WBOYOriginal
2016-07-21 14:53:151856Durchsuche

声明

本文基于Python2.7语言,给出判断列表是否已排序的多种方法,并在作者的Windows XP主机(Pentium G630 2.7GHz主频2GB内存)上对比和分析其性能表现。

一. 问题提出

Haskell培训老师提出一个问题:如何判断列表是否已经排序?

排序与否实际只是相邻元素间的某种二元关系,即a->a->Bool。所以第一步可以把二元组列表找出来;第二步是把这个函数作用于每个元组,然后用and操作。老师给出的实现代码如下:

pair lst = zip lst ( tail lst )
sorted lst predict = and [ predict x y | (x,y) <- pair lst] 

Haskell中,等号前面是函数的名称和参数,后面是函数的定义体。pair函数将列表lst错位一下(tail除去列表的第一个元素)后,和原列表在zip的作用下形成前后相邻元素二元组列表。predict函数接受两个数字,根据大小返回True或False。and对类型为[Bool]的列表中所有元素求与,其语义类似Python的all()函数。

随后,老师请大家思考性能问题。作者提出,若准确性要求不高,可生成三组随机数排序后作为下标,提取原列表相应的三组元素组成三个新的子列表("采样")。若判断三个子列表遵循同样的排序规则时,则认为原列表已排序。当列表很大且前段已排序时,选取适当数目的随机数,可在保障一定准确率的同时显著地降低运算规模。

此外,实际应用中还应考虑数据来源。例如,Python语言的os.listdir()方法在Windows系统中返回的列表条目满足字母序,在Linux系统中则返回乱序条目。因此,若判断系统平台(os.platform)为win32,则条目必然已经排序。

为对比验证随机采样方式的可行性,作者先在网上搜集判断列表排序的现有方法,主要参考Stack Overflow网站上"Pythonic way to check if a list is sorted or not"问题的答案,并在本文第二节给出相关的代码示例。注意,本文所述的"排序"不要求严格排序,即相邻元素允许相等。例如,[1,2,2,3]视为升序,[3,3,2,2]视为降序。

二. 代码实现

本节判断列表排序的函数名格式为IsListSorted_XXX()。为简洁起见,除代码片段及其输出外,一律以_XXX()指代。

2.1 guess

def IsListSorted_guess(lst):
listLen = len(lst)
if listLen <= 1:
return True
#由首个元素和末尾元素猜测可能的排序规则
if lst[0] == lst[-1]: #列表元素相同
for elem in lst:
if elem != lst[0]: return False
elif lst[0] < lst[-1]: #列表元素升序
for i, elem in enumerate(lst[1:]):
if elem < lst[i]: return False
else: #列表元素降序
for i, elem in enumerate(lst[1:]):
if elem > lst[i]: return False
return True 

_guess()是最通用的实现,几乎与语言无关。值得注意的是,该函数内会猜测给定列表可能的排序规则,因此无需外部调用者指明排序规则。

2.2 sorted

def IsListSorted_sorted(lst):
return sorted(lst) == lst or sorted(lst, reverse=True) == lst 

_sorted()使用Python内置函数sorted()。由于sorted()会对未排序的列表排序,_sorted()函数主要适用于已排序列表。
若想判断列表未排序后再对其排序,不如直接调用列表的sort()方法,因为该方法内部会判断列表是否排序。对于已排序列表,该方法的时间复杂度为线性阶O(n)——判断为O(n)而排序为O(nlgn)。

2.3 for-loop

def IsListSorted_forloop(lst, key=lambda x, y: x <= y):
for i, elem in enumerate(lst[1:]): #注意,enumerate默认迭代下标从0开始
if not key(lst[i], elem): #if elem > lst[i]更快,但通用性差
return False
return True 

无论列表是否已排序,本函数的时间复杂度均为线性阶O(n)。注意,参数key表明缺省的排序规则为升序。

2.4 all

def IsListSorted_allenumk(lst, key=lambda x, y: x <= y):
return all(key(lst[i], elem) for i, elem in enumerate(lst[1:]))
import operator
def IsListSorted_allenumo(lst, oCmp=operator.le):
return all(oCmp(lst[i], elem) for i, elem in enumerate(lst[1:]))
def IsListSorted_allenumd(lst):
return all((lst[i] <= elem) for i, elem in enumerate(lst[1:]))
def IsListSorted_allxran(lst, key=lambda x,y: x <= y):
return all(key(lst[i],lst[i+1]) for i in xrange(len(lst)-1))
def IsListSorted_allzip(lst, key=lambda x,y: x <= y):
from itertools import izip #Python 3中zip返回生成器(generator),而izip被废弃
return all(key(a, b) for (a, b) in izip(lst[:-1],lst[1:])) 

lambda表达式与operator运算符速度相当,前者简单灵活,后者略为高效(实测并不一定)。但两者速度均不如列表元素直接比较(可能存在调用开销)。亦即,_allenumd()快于_allenumo()快于_allenumk()。

若使用lambda表达式指示排序规则,更改规则时只需要改变x和y之间的比较运算符;若使用operator模块指示排序规则,更改规则时需要改变对象比较方法。具体地,lt(x, y)等效于x y,ge(x, y)等效于x >= y。例如,_allenumo()函数若要严格升序可设置oCmp=operator.lt。

此外,由all()函数的帮助信息可知,_allenumk()其实是_forloop()的等效形式。

2.5 numpy

def IsListSorted_numpy(arr, key=lambda dif: dif >= 0):
import numpy
try:
if arr.dtype.kind == 'u': #无符号整数数组执行np.diff时存在underflow风险
arr = numpy.int64(lst)
except AttributeError:
pass #无dtype属性,非数组
return (key(numpy.diff(arr))).all() #numpy.diff(x)返回相邻数组元素的差值构成的数组 

NumPy是用于科学计算的Python基础包,可存储和处理大型矩阵。它包含一个强大的N维数组对象,比Python自身的嵌套列表结构(nested list structure)高效得多。第三节的实测数据表明,_numpy()处理大型列表时性能非常出色。

在Windows系统中可通过pip install numpy命令安装NumPy包,不建议登录官网下载文件自行安装。

2.6 reduce

def IsListSorted_reduce(iterable, key=lambda x, y: x <= y):
cmpFunc = lambda x, y: y if key(x, y) else float('inf')
return reduce(cmpFunc, iterable, .0) < float('inf') 

reduce实现是all实现的变体。累加器(accumulator)中仅存储最后一个检查的列表元素,或者Infinity(若任一元素小于前个元素值)。

前面2.1~2.5小节涉及下标操作的函数适用于列表等可迭代对象(Iterable)。对于通用迭代器(Iterator)对象,即可以作用于next()函数或方法的对象,可使用_reduce()及后面除_rand()外各小节的函数。迭代器的计算是惰性的,只有在需要返回下一个数据时才会计算,以避免不必要的计算。而且,迭代器方式无需像列表那样切片为两个迭代对象。

2.7 imap

def IsListSorted_itermap(iterable, key=lambda x, y: x <= y):
from itertools import imap, tee
a, b = tee(iterable) #为单个iterable创建两个独立的iterator
next(b, None)
return all(imap(key, a, b)) 

2.8 izip

def IsListSorted_iterzip(iterable, key=lambda x, y: x <= y):
from itertools import tee, izip
a, b = tee(iterable)
next(b, None)
return all(key(x, y) for x, y in izip(a, b))
def pairwise(iterable):
from itertools import tee, izip
a, b = tee(iterable)
next(b, None)
return izip(a, b) #"s -> (s0,s1), (s1,s2), (s2, s3), ..."
def IsListSorted_iterzipf(iterable, key=lambda x, y: x <= y):
return all(key(a, b) for a, b in pairwise(iterable)) 

第三节的实测数据表明,虽然存在外部函数调用,_iterzipf()却比_iterzip()略为高效。

2.9 fast

def IsListSorted_fastd(lst):
it = iter(lst)
try:
prev = it.next()
except StopIteration:
return True
for cur in it:
if prev > cur:
return False
prev = cur
return True
def IsListSorted_fastk(lst, key=lambda x, y: x <= y):
it = iter(lst)
try:
prev = it.next()
except StopIteration:
return True
for cur in it:
if not key(prev, cur):
return False
prev = cur
return True 

_fastd()和_fastk()是Stack Overflow网站回答里据称执行最快的。实测数据表明,在列表未排序时,它们的性能表现确实优异。

2.10 random

import random
def IsListSorted_rand(lst, randNum=3, randLen=100):
listLen = len(lst)
if listLen <= 1:
return True

#由首个元素和末尾元素猜测可能的排序规则
if lst[0] < lst[-1]: #列表元素升序
key = lambda dif: dif >= 0
else: #列表元素降序或相等
key = lambda dif: dif <= 0

threshold, sortedFlag = 10000, True
import numpy
if listLen <= threshold or listLen <= randLen*2 or not randNum:
return (key(numpy.diff(numpy.array(lst)))).all()
from random import sample
for i in range(randNum):
sortedRandList = sorted(sample(xrange(listLen), randLen))
flag = (key(numpy.diff(numpy.array([lst[x] for x in sortedRandList])))).all()
sortedFlag = sortedFlag and flag
return sortedFlag 

_rand()借助随机采样降低运算规模,并融入其他判断函数的优点。例如,猜测列表可能的排序规则,并在随机采样不适合时使用相对快速的判断方式,如NumPy。

通过line_profiler分析可知,第20行和第21行与randLen有关,但两者耗时接近。因此randLen应小于listLen的一半,以抵消sorted开销。除内部限制外,用户可以调节随机序列个数和长度,如定制单个但较长的序列。

注意,_rand()不适用于存在微量异常数据的长列表。因为这些数据很可能被随机采样遗漏,从而影响判断结果的准确性。

三. 性能分析

本节借助Python标准库random模块,生成各种随机列表,以对比和分析上节列表排序判断函数的性能。

可通过sample()、randint()等方法生成随机列表。例如:

>>>import random
>>> random.sample(range(10), 10); random.sample(range(10), 5)
[9, 1, 6, 3, 0, 8, 4, 2, 7, 5]
[1, 2, 5, 6, 0]
>>> rand = [random.randint(1,10) for i in range(10)]; rand
[7, 3, 7, 5, 7, 2, 4, 4, 9, 8]
>>> random.sample(rand, 5); random.sample(rand, 5)
[4, 7, 7, 9, 8]
[3, 9, 2, 5, 7]
>>> randGen = (random.randint(1,10) for i in range(10)); randGen
<generator object <genexpr> at 0x0192C878>

sample()方法从列表中随机选择数字,结合range()函数可生产不含重复元素的随机列表;而randint()方法结合列表解析生成的随机列表可能包含重复元素。Python文档规定,首次导入random模块时使用当前系统时间作为种子初始化随机数生成器。因此,本文并未显式地调用seed()方法设置种子。

为度量性能表现,定义如下计时装饰器:

from time import clock
TimeList = []
def FuncTimer(repeats=1000):
def decorator(func):
def wrapper(*args, **kwargs):
try:
startTime = clock()
for i in xrange(repeats):
ret = func(*args, **kwargs)
except Exception, e:
print '%s() Error: %s!' %(func.__name__, e)
ret = None
finally: #当目标函数发生异常时,仍旧输出计时信息
endTime = clock()
timeElasped = (endTime-startTime)*1000.0
msg = '%21s(): %s =>Time Elasped: %.3f msec, repeated %d time(s).' \
%(func.__name__, ret, timeElasped, repeats)
global TimeList; TimeList.append([timeElasped, msg])
return ret
return wrapper
return decorator
def ReportTimer():
global TimeList; TimeList.sort(key=lambda x:x[0])
for entry in TimeList:
print entry[1]
TimeList = [] #Flush

该装饰器允许对输出进行排序,以便更直观地观察性能。基于该装饰器,下文将分别测试不同的排序场景。注意,第二节各函数头部需添加FuncTimer()装饰。

3.1 列表前段乱序

测试代码如下:

def TestHeadUnorderedList():
TEST_NAME = 'HeadUnorderedList'; scale = int(1e5)
List = random.sample(xrange(scale), scale) + range(scale)
print 'Test 1: %s, list len: %d' %(TEST_NAME, len(List))
IsListSorted_guess(List)
IsListSorted_sorted(List)
IsListSorted_allenumk(List)
IsListSorted_allenumo(List)
IsListSorted_allenumd(List)
IsListSorted_allxran(List)
IsListSorted_allzip(List)
IsListSorted_forloop(List)
IsListSorted_itermap(List)
IsListSorted_iterzipf(List)
IsListSorted_iterzip(List)
IsListSorted_reduce(List)
IsListSorted_numpy(numpy.array(List)) #若不先转换为数组,则耗时骤增
IsListSorted_fastd(List)
IsListSorted_fastk(List)
ReportTimer()

运行输出如下:

Test 1: HeadUnorderedList, list len: 200
IsListSorted_fastd(): False =>Time Elasped: 0.757 msec, repeated 1000 time(s).
IsListSorted_fastk(): False =>Time Elasped: 1.091 msec, repeated 1000 time(s).
IsListSorted_forloop(): False =>Time Elasped: 2.080 msec, repeated 1000 time(s).
IsListSorted_guess(): False =>Time Elasped: 2.123 msec, repeated 1000 time(s).
IsListSorted_allxran(): False =>Time Elasped: 2.255 msec, repeated 1000 time(s).
IsListSorted_allenumd(): False =>Time Elasped: 2.672 msec, repeated 1000 time(s).
IsListSorted_allenumo(): False =>Time Elasped: 3.021 msec, repeated 1000 time(s).
IsListSorted_allenumk(): False =>Time Elasped: 3.207 msec, repeated 1000 time(s).
IsListSorted_itermap(): False =>Time Elasped: 5.845 msec, repeated 1000 time(s).
IsListSorted_allzip(): False =>Time Elasped: 7.793 msec, repeated 1000 time(s).
IsListSorted_iterzip(): False =>Time Elasped: 9.667 msec, repeated 1000 time(s).
IsListSorted_iterzipf(): False =>Time Elasped: 9.969 msec, repeated 1000 time(s).
IsListSorted_numpy(): False =>Time Elasped: 16.203 msec, repeated 1000 time(s).
IsListSorted_sorted(): False =>Time Elasped: 45.742 msec, repeated 1000 time(s).
IsListSorted_reduce(): False =>Time Elasped: 145.447 msec, repeated 1000 time(s).
Test 1: HeadUnorderedList, list len: 200000
IsListSorted_fastd(): False =>Time Elasped: 0.717 msec, repeated 1000 time(s).
IsListSorted_fastk(): False =>Time Elasped: 0.876 msec, repeated 1000 time(s).
IsListSorted_allxran(): False =>Time Elasped: 2.104 msec, repeated 1000 time(s).
IsListSorted_itermap(): False =>Time Elasped: 6.062 msec, repeated 1000 time(s).
IsListSorted_iterzip(): False =>Time Elasped: 7.244 msec, repeated 1000 time(s).
IsListSorted_iterzipf(): False =>Time Elasped: 8.491 msec, repeated 1000 time(s).
IsListSorted_numpy(): False =>Time Elasped: 801.916 msec, repeated 1000 time(s).
IsListSorted_forloop(): False =>Time Elasped: 2924.755 msec, repeated 1000 time(s).
IsListSorted_guess(): False =>Time Elasped: 2991.756 msec, repeated 1000 time(s).
IsListSorted_allenumo(): False =>Time Elasped: 3025.864 msec, repeated 1000 time(s).
IsListSorted_allenumk(): False =>Time Elasped: 3062.792 msec, repeated 1000 time(s).
IsListSorted_allenumd(): False =>Time Elasped: 3190.896 msec, repeated 1000 time(s).
IsListSorted_allzip(): False =>Time Elasped: 6586.183 msec, repeated 1000 time(s).
IsListSorted_sorted(): False =>Time Elasped: 119974.955 msec, repeated 1000 time(s).
IsListSorted_reduce(): False =>Time Elasped: 154747.659 msec, repeated 1000 time(s).

可见,对于前段乱序的列表,无论其长短_fastd()和_fastk()的表现均为最佳。对于未排序列表,_sorted()需要进行排序,故性能非常差。然而,_reduce()性能最差。

实际上除_guess()和_sorted()外,其他函数都按升序检查列表。为安全起见,可仿照_guess()实现,先猜测排序方式,再进一步检查。

因为短列表耗时差异大多可以忽略,后续测试将不再包含短列表(但作者确实测试过),仅关注长列表。除非特别说明,列表长度为10万级,重复计时1000次。

3.2 列表中段乱序

测试代码及结果如下:

def TestMiddUnorderedList():
TEST_NAME = 'MiddUnorderedList'; scale = int(1e5)
List = range(scale) + random.sample(xrange(scale), scale) + range(scale)
print 'Test 2: %s, list len: %d' %(TEST_NAME, len(List))
IsListSorted_numpy(numpy.array(List)) # 1572.295 msec 
IsListSorted_guess(List) # 14540.637 msec
IsListSorted_itermap(List) # 21013.096 msec
IsListSorted_fastk(List) # 23840.582 msec
IsListSorted_allxran(List) # 31014.215 msec
IsListSorted_iterzip(List) # 33386.059 msec
IsListSorted_forloop(List) # 34228.006 msec
IsListSorted_allenumk(List) # 34416.802 msec
IsListSorted_allzip(List) # 42370.528 msec
IsListSorted_sorted(List) # 142592.756 msec
IsListSorted_reduce(List) # 187514.967 msec
ReportTimer()

为节省篇幅,已根据运行输出调整函数的调用顺序。下文也作如此处理。

3.3 列表后段乱序

测试代码及结果如下:

def TestTailUnorderedList():
TEST_NAME = 'TailUnorderedList'; scale = int(1e5)
List = range(scale, 0, -1) + random.sample(xrange(scale), scale)
print 'Test 3: %s, list len: %d' %(TEST_NAME, len(List))
IsListSorted_numpy(numpy.array(List), key=lambda dif: dif <= 0) # 980.789 msec 
IsListSorted_guess(List) # 13273.862 msec
IsListSorted_itermap(List, key=lambda x, y: x >= y) # 21603.315 msec
IsListSorted_fastk(List, key=lambda x, y: x >= y) # 24183.548 msec
IsListSorted_allxran(List, key=lambda x, y: x >= y) # 32850.254 msec
IsListSorted_forloop(List, key=lambda x, y: x >= y) # 33918.848 msec
IsListSorted_iterzip(List, key=lambda x, y: x >= y) # 34505.809 msec
IsListSorted_allenumk(List, key=lambda x, y: x >= y) # 35631.625 msec
IsListSorted_allzip(List, key=lambda x, y: x >= y) # 40076.330 msec
ReportTimer()

3.4 列表完全乱序

测试代码及结果如下:

def TestUnorderedList():
TEST_NAME = 'UnorderedList'; scale = int(1e5)
List = random.sample(xrange(scale), scale)
print 'Test 4: %s, list len: %d' %(TEST_NAME, len(List))
IsListSorted_fastk(List) # 0.856 msec
IsListSorted_allxran(List) # 3.438 msec
IsListSorted_iterzip(List) # 7.233 msec
IsListSorted_itermap(List) # 7.595 msec
IsListSorted_numpy(numpy.array(List)) # 207.222 msec
IsListSorted_allenumk(List) # 953.423 msec
IsListSorted_guess(List) # 1023.575 msec
IsListSorted_forloop(List) # 1076.986 msec
IsListSorted_allzip(List) # 2062.821 msec
ReportTimer()

3.5 列表元素相同

测试代码及结果如下:

```python def TestSameElemList(): TEST_NAME = 'SameElemList'; scale = int(1e5) List = [5]*scale print 'Test 5: %s, list len: %d' %(TEST_NAME, len(List)) IsListSorted_numpy(numpy.array(List)) # 209.324 msec IsListSorted_sorted(List) # 2760.139 msec IsListSorted_guess(List) # 5843.942 msec IsListSorted_itermap(List) # 20609.704 msec IsListSorted_fastk(List) # 23035.760 msec IsListSorted_forloop(List) # 29043.206 msec IsListSorted_allenumk(List) # 29553.716 msec IsListSorted_allxran(List) # 30348.549 msec IsListSorted_iterzip(List) # 32806.217 msec ReportTimer()

3.6 列表升序

测试代码及结果如下:

def TestAscendingList():
TEST_NAME = 'AscendingList'; scale = int(1e5)
List = range(scale)
print 'Test 6: %s, list len: %d' %(TEST_NAME, len(List))
IsListSorted_numpy(numpy.array(List)) # 209.217 msec
IsListSorted_sorted(List) # 2845.166 msec
IsListSorted_fastd(List) # 5977.520 msec
IsListSorted_guess(List) # 10408.204 msec
IsListSorted_allenumd(List) # 15812.754 msec
IsListSorted_itermap(List) # 21244.476 msec
IsListSorted_fastk(List) # 23900.196 msec
IsListSorted_allenumo(List) # 28607.724 msec
IsListSorted_forloop(List) # 30075.538 msec
IsListSorted_allenumk(List) # 30274.017 msec
IsListSorted_allxran(List) # 31126.404 msec
IsListSorted_reduce(List) # 32940.108 msec
IsListSorted_iterzip(List) # 34188.312 msec
IsListSorted_iterzipf(List) # 34425.097 msec
IsListSorted_allzip(List) # 37967.447 msec
ReportTimer()

可见,列表已排序时,_sorted()的性能较占优势。

3.7 列表降序

剔除不支持降序的函数,测试代码及结果如下:

def TestDescendingList():
TEST_NAME = 'DescendingList'; scale = int(1e2)
List = range(scale, 0, -1)
print 'Test 7: %s, list len: %d' %(TEST_NAME, len(List))
IsListSorted_numpy(numpy.array(List), key=lambda dif: dif <= 0) # 209.318 msec
IsListSorted_sorted(List) # 5707.067 msec
IsListSorted_guess(List) # 10549.928 msec
IsListSorted_itermap(List, key=lambda x, y: x >= y) # 21529.547 msec
IsListSorted_fastk(List, key=lambda x, y: x >= y) # 24264.465 msec
import operator; IsListSorted_allenumo(List, oCmp=operator.ge) # 28093.035 msec
IsListSorted_forloop(List, key=lambda x, y: x >= y) # 30745.943 msec
IsListSorted_allenumk(List, key=lambda x, y: x >= y) # 32696.205 msec
IsListSorted_allxran(List, key=lambda x, y: x >= y) # 33431.473 msec
IsListSorted_allzip(List, key=lambda x, y: x >= y) # 34837.019 msec
IsListSorted_iterzip(List, key=lambda x, y: x >= y) # 35237.475 msec
IsListSorted_reduce(List, key=lambda x, y: x >= y) # 37035.270 msec
ReportTimer()

3.8 迭代器测试

参数为列表的函数,需要先将列表���过iter()函数转换为迭代器。注意,当iterable参数为iterator时,只能计时一次,因为该次执行将用尽迭代器。

测试代码如下:

def TestIter():
TEST_NAME = 'Iter'; scale = int(1e7)
List = range(scale) #升序
# List = random.sample(xrange(scale), scale) #乱序
print 'Test 8: %s, list len: %d' %(TEST_NAME, len(List))
iterL = iter(List); IsListSorted_guess(list(iterL))
iterL = iter(List); IsListSorted_sorted(iterL)
iterL = iter(List); IsListSorted_itermap(iterL)
iterL = iter(List); IsListSorted_iterzipf(iterL)
iterL = iter(List); IsListSorted_iterzip(iterL)
iterL = iter(List); IsListSorted_reduce(iterL)
iterL = iter(List); IsListSorted_fastd(iterL)
iterL = iter(List); IsListSorted_fastk(iterL, key=lambda x, y: x >= y)
ReportTimer()

运行结果如下:

Test 8: Iter, list len: 10000000 ---升序
IsListSorted_fastd(): True =>Time Elasped: 611.028 msec, repeated 1 time(s).
IsListSorted_sorted(): False =>Time Elasped: 721.751 msec, repeated 1 time(s).
IsListSorted_guess(): True =>Time Elasped: 1142.065 msec, repeated 1 time(s).
IsListSorted_itermap(): True =>Time Elasped: 2097.696 msec, repeated 1 time(s).
IsListSorted_fastk(): True =>Time Elasped: 2337.233 msec, repeated 1 time(s).
IsListSorted_reduce(): True =>Time Elasped: 3307.361 msec, repeated 1 time(s).
IsListSorted_iterzipf(): True =>Time Elasped: 3354.034 msec, repeated 1 time(s).
IsListSorted_iterzip(): True =>Time Elasped: 3442.520 msec, repeated 1 time(s).
Test 8: Iter, list len: 10000000 ---乱序
IsListSorted_fastk(): False =>Time Elasped: 0.004 msec, repeated 1 time(s).
IsListSorted_fastd(): False =>Time Elasped: 0.010 msec, repeated 1 time(s).
IsListSorted_iterzip(): False =>Time Elasped: 0.015 msec, repeated 1 time(s).
IsListSorted_itermap(): False =>Time Elasped: 0.055 msec, repeated 1 time(s).
IsListSorted_iterzipf(): False =>Time Elasped: 0.062 msec, repeated 1 time(s).
IsListSorted_guess(): False =>Time Elasped: 736.810 msec, repeated 1 time(s).
IsListSorted_reduce(): False =>Time Elasped: 8919.611 msec, repeated 1 time(s).
IsListSorted_sorted(): False =>Time Elasped: 12273.018 msec, repeated 1 time(s).

其中,_itermap()、_iterzip()、_iterzipf()、_reduce()、_fastd()、_fastk()可直接判断迭代器是否已排序。其他函数需将迭代器转换为列表后才能处理。_sorted()虽然接受迭代器参数,但sorted()返回列表,故无法正确判断迭代器顺序。

3.9 随机采样测试

综合以上测试,可知_fastk()和_numpy()性能较为突出,而且_rand()内置numpy方式。因此,以_fastk()和_numpy()为参照对象,测试代码如下(计时1次):

def TestRandList():
scale = int(1e6)
List = random.sample(xrange(scale), scale) + range(scale)
print 'Test 1: %s, list len: %d' %('HeadUnorderedList', len(List))
IsListSorted_fastk(List)
IsListSorted_numpy(numpy.array(List))
IsListSorted_rand(List, randNum=1)
ReportTimer()
List = range(scale) + random.sample(xrange(scale), scale) + range(scale)
print 'Test 2: %s, list len: %d' %('MiddUnorderedList', len(List))
IsListSorted_fastk(List)
IsListSorted_numpy(numpy.array(List))
IsListSorted_rand(List, randNum=1)
ReportTimer()
List = range(scale, 0, -1) + random.sample(xrange(scale), scale)
print 'Test 3: %s, list len: %d' %('TailUnorderedList', len(List))
IsListSorted_fastk(List, key=lambda x, y: x >= y)
IsListSorted_numpy(numpy.array(List), key=lambda dif: dif <= 0)
IsListSorted_rand(List, randNum=1)
ReportTimer()
List = [random.randint(1,scale) for i in xrange(scale)] #random.sample(xrange(scale), scale)
print 'Test 4: %s, list len: %d' %('UnorderedList', len(List))
IsListSorted_fastk(List)
IsListSorted_numpy(numpy.array(List))
IsListSorted_rand(List, randNum=1)
ReportTimer()
List = [5]*scale
print 'Test 5: %s, list len: %d' %('SameElemList', len(List))
IsListSorted_fastk(List)
IsListSorted_numpy(numpy.array(List))
IsListSorted_rand(List, randNum=1)
ReportTimer()
List = range(scale)
print 'Test 6: %s, list len: %d' %('AscendingList', len(List))
IsListSorted_fastk(List)
IsListSorted_numpy(numpy.array(List))
IsListSorted_rand(List, randNum=1)
ReportTimer()
List = range(scale, 0, -1)
print 'Test 7: %s, list len: %d' %('DescendingList', len(List))
IsListSorted_fastk(List, key=lambda x, y: x >= y)
IsListSorted_numpy(numpy.array(List), key=lambda dif: dif <= 0)
IsListSorted_rand(List, randNum=1)
ReportTimer()
List = range(scale, 0, -1); List[scale/2]=0
print 'Test 8: %s, list len: %d' %('MiddleNotchList', len(List))
IsListSorted_fastk(List, key=lambda x, y: x >= y)
IsListSorted_numpy(numpy.array(List), key=lambda dif: dif <= 0)
IsListSorted_rand(List, randNum=1)
IsListSorted_rand(List, randNum=1, randLen=scale/2)
ReportTimer()

运行输出如下:

Test 1: HeadUnorderedList, list len: 2000000
IsListSorted_fastk(): False =>Time Elasped: 0.095 msec, repeated 1 time(s).
IsListSorted_rand(): False =>Time Elasped: 0.347 msec, repeated 1 time(s).
IsListSorted_numpy(): False =>Time Elasped: 7.893 msec, repeated 1 time(s).
Test 2: MiddUnorderedList, list len: 3000000
IsListSorted_rand(): False =>Time Elasped: 0.427 msec, repeated 1 time(s).
IsListSorted_numpy(): False =>Time Elasped: 11.868 msec, repeated 1 time(s).
IsListSorted_fastk(): False =>Time Elasped: 210.842 msec, repeated 1 time(s).
Test 3: TailUnorderedList, list len: 2000000
IsListSorted_rand(): False =>Time Elasped: 0.355 msec, repeated 1 time(s).
IsListSorted_numpy(): False =>Time Elasped: 7.548 msec, repeated 1 time(s).
IsListSorted_fastk(): False =>Time Elasped: 280.416 msec, repeated 1 time(s).
Test 4: UnorderedList, list len: 1000000
IsListSorted_fastk(): False =>Time Elasped: 0.074 msec, repeated 1 time(s).
IsListSorted_rand(): False =>Time Elasped: 0.388 msec, repeated 1 time(s).
IsListSorted_numpy(): False =>Time Elasped: 3.757 msec, repeated 1 time(s).
Test 5: SameElemList, list len: 1000000
IsListSorted_rand(): True =>Time Elasped: 0.304 msec, repeated 1 time(s).
IsListSorted_numpy(): True =>Time Elasped: 3.955 msec, repeated 1 time(s).
IsListSorted_fastk(): True =>Time Elasped: 210.977 msec, repeated 1 time(s).
Test 6: AscendingList, list len: 1000000
IsListSorted_rand(): True =>Time Elasped: 0.299 msec, repeated 1 time(s).
IsListSorted_numpy(): True =>Time Elasped: 4.822 msec, repeated 1 time(s).
IsListSorted_fastk(): True =>Time Elasped: 214.171 msec, repeated 1 time(s).
Test 7: DescendingList, list len: 1000000
IsListSorted_rand(): True =>Time Elasped: 0.336 msec, repeated 1 time(s).
IsListSorted_numpy(): True =>Time Elasped: 3.867 msec, repeated 1 time(s).
IsListSorted_fastk(): True =>Time Elasped: 279.322 msec, repeated 1 time(s).
Test 8: MiddleNotchList, list len: 1000000
IsListSorted_rand(): True =>Time Elasped: 0.387 msec, repeated 1 time(s).
IsListSorted_numpy(): False =>Time Elasped: 4.792 msec, repeated 1 time(s).
IsListSorted_rand(): False =>Time Elasped: 78.903 msec, repeated 1 time(s).
IsListSorted_fastk(): False =>Time Elasped: 110.444 msec, repeated 1 time(s).

可见,在绝大部分测试场景中,_rand()性能均为最佳,且不失正确率。注意测试8,当降序列表中间某个元素被置0(开槽)时,随机采样很容易遗漏该元素,导致误判。然而,这种场景在实际使用中非常罕见。

Stellungnahme:
Der Inhalt dieses Artikels wird freiwillig von Internetnutzern beigesteuert und das Urheberrecht liegt beim ursprünglichen Autor. Diese Website übernimmt keine entsprechende rechtliche Verantwortung. Wenn Sie Inhalte finden, bei denen der Verdacht eines Plagiats oder einer Rechtsverletzung besteht, wenden Sie sich bitte an admin@php.cn