代码与图详解性能之Python集合类型(list tuple dict set generator)

时间:2022-01-12 02:12:44
Python内嵌的集合类型有list、tuple、set、dict。
列表list:看似数组,但比数组强大,支持索引、切片、查找、增加等功能。
元组tuple:功能跟list差不多,但一旦生成,长度及元素都不可变(元素的元素还是可变),似乎就是一更轻量级、安全的list。
字典dict:键值对结构哈希表,跟哈希表的性质一样,key无序且不重复,增删改方便快捷。
set:无序且不重复的集合,就是一个只有键没有值的dict,Java的HashSet就是采用HashMap实现,但愿python不会是这样,毕竟set不需要value,省去了很多指针。
 
称之为生成器,或者列表推导式,是python中有一个特殊的数据类型,实际上并不是一个数据结构,只包括算法和暂存的状态,并且具有迭代的功能。
 
先看看它们的内存使用情况,分别用生成器生成100000个元素的set, dict, generator, tuple, list。消耗的内存dict, set, list, tuple依次减少,生成的对象大小也是一样。由于generator并不生成数据表,所以不需要消耗内存:
import sys
from memory_profiler import profile @profile
def create_data(data_size):
data_generator = (x for x in xrange(data_size))
data_set = {x for x in xrange(data_size)}
data_dict = {x:None for x in xrange(data_size)}
data_tuple = tuple(x for x in xrange(data_size))
data_list = [x for x in xrange(data_size)]
return data_set, data_dict, data_generator, data_tuple, data_list data_size = 100000
for data in create_data(data_size):
print data.__class__, sys.getsizeof(data) Line # Mem usage Increment Line Contents
================================================
4 14.6 MiB 0.0 MiB @profile
5 def create_data(data_size):
6 14.7 MiB 0.0 MiB data_generator = (x for x in xrange(data_size))
7 21.4 MiB 6.7 MiB data_set = {x for x in xrange(data_size)}
8 29.8 MiB 8.5 MiB data_dict = {x:None for x in xrange(data_size)}
9 33.4 MiB 3.6 MiB data_tuple = tuple(x for x in xrange(data_size))
10 38.2 MiB 4.8 MiB data_list = [x for x in xrange(data_size)]
11 38.2 MiB 0.0 MiB return data_set, data_dict, data_generator, data_tuple, data_list <type 'set'> 4194528
<type 'dict'> 6291728
<type 'generator'> 72
<type 'tuple'> 800048
<type 'list'> 824464

再看看查找性能,dict,set是常数查找时间(O(1)),list、tuple是线性查找时间(O(n)),用生成器生成指定大小元素的对象,用随机生成的数字去查找:

import time
import sys
import random
from memory_profiler import profile def create_data(data_size):
data_set = {x for x in xrange(data_size)}
data_dict = {x:None for x in xrange(data_size)}
data_tuple = tuple(x for x in xrange(data_size))
data_list = [x for x in xrange(data_size)]
return data_set, data_dict, data_tuple, data_list def cost_time(func):
def cost(*args, **kwargs):
start = time.time()
r = func(*args, **kwargs)
cost = time.time() - start
print 'find in %s cost time %s' % (r, cost)
return r, cost #返回数据的类型和方法执行消耗的时间
return cost @cost_time
def test_find(test_data, data):
for d in test_data:
if d in data:
pass
return data.__class__.__name__ data_size = 100
test_size = 10000000
test_data = [random.randint(0, data_size) for x in xrange(test_size)]
#print test_data
for data in create_data(data_size):
test_find(test_data, data) 输出:
----------------------------------------------
find in <type 'set'> cost time 0.47200012207
find in <type 'dict'> cost time 0.429999828339
find in <type 'tuple'> cost time 5.36500000954
find in <type 'list'> cost time 5.53399991989
100个元素的大小的集合,分别查找1000W次,差距非常明显。不过这些随机数,都是能在集合中查找得到。修改一下随机数方式,生成一半是能查找得到,一半是查找不到的。从打印信息可以看出在有一半最坏查找例子的情况下,list、tuple表现得更差了。
def randint(index, data_size):
return random.randint(0, data_size) if (x % 2) == 0 else random.randint(data_size, data_size * 2) test_data = [randint(x, data_size) for x in xrange(test_size)] 输出:
----------------------------------------------
find in <type 'set'> cost time 0.450000047684
find in <type 'dict'> cost time 0.397000074387
find in <type 'tuple'> cost time 7.83299994469
find in <type 'list'> cost time 8.27800011635

元素的个数从10增长至500,统计每次查找10W次的时间,用图拟合时间消耗的曲线,结果如下图,结果证明dict, set不管元素多少,一直都是常数查找时间,dict、tuple随着元素增长,呈现线性增长时间:

import matplotlib.pyplot as plot
from numpy import * data_size = array([x for x in xrange(10, 500, 10)])
test_size = 100000
cost_result = {}
for size in data_size:
test_data = [randint(x, size) for x in xrange(test_size)]
for data in create_data(size):
name, cost = test_find(test_data, data) #装饰器函数返回函数的执行时间
cost_result.setdefault(name, []).append(cost) plot.figure(figsize=(10, 6))
xline = data_size
for data_type, result in cost_result.items():
yline = array(result)
plot.plot(xline, yline, label=data_type) plot.ylabel('Time spend')
plot.xlabel('Find times') plot.grid() plot.legend()
plot.show()

代码与图详解性能之Python集合类型(list tuple dict set generator)

迭代的时间,区别很微弱,dict、set要略微消耗时间多一点:

@cost_time
def test_iter(data):
for d in data:
pass
return data.__class__ .__name__ data_size = array([x for x in xrange(1, 500000, 1000)])
cost_result = {}
for size in data_size:
for data in create_data(size):
name, cost = test_iter(data)
cost_result.setdefault(name, []).append(cost) #拟合曲线图
plot.figure(figsize=(10, 6))
xline = data_size
for data_type, result in cost_result.items():
yline = array(result)
plot.plot(xline, yline, label=data_type) plot.ylabel('Time spend')
plot.xlabel('Iter times') plot.grid() plot.legend()
plot.show()

代码与图详解性能之Python集合类型(list tuple dict set generator)

删除元素消耗时间图示如下,随机删除1000个元素,tuple类型不能删除元素,所以不做比较:

@cost_time
def test_delete(test_data, data):
for d in test_data:
data.remove(d)
return data.__class__.__name__ @cost_time
def test_dict_delete(test_data, data):
for d in test_data:
del data[d]
return data.__class__.__name__ def create_data(data_size):
data_set = {x for x in xrange(data_size)}
data_dict = {x:None for x in xrange(data_size)}
data_list = [x for x in xrange(data_size)]
return data_set, data_dict, data_list #创建随机删除数据集
def create_random_test_data(size, range_size):
test_data = set()
while(len(test_data) < size):
test_data.add(random.randint(0, range_size))
return test_data #dict没有remove方法,用del dict[key]来删除数据,其他数据类型使用remove方法
delete_method = {list: test_delete, set: test_delete, dict: test_dict_delete} #每次检测1000增量大小的数据的删除一半时间
data_size = array([x for x in xrange(1000, 20000, 1000)])
cost_result = {}
test_size = 1000
for size in data_size:
test_data = create_random_test_data(test_size, size)
for data in create_data(size):
name, cost = delete_method[type(data)](test_data, data) #返回数据类型的名字和方法的执行时间
cost_result.setdefault(name, []).append(cost)

代码与图详解性能之Python集合类型(list tuple dict set generator)

随机删除一半的元素,图形就呈指数时间(O(n2))增长了:

代码与图详解性能之Python集合类型(list tuple dict set generator)

添加元素消耗的时间图示如下,统计以10000为增量大小的元素个数的添加时间,都是线性增长时间,看不出有什么差别,tuple类型不能添加新的元素,所以不做比较:
@cost_time
def test_dict_add(test_data, data):
for d in test_data:
data[d] = None
return data.__class__ .__name__ @cost_time
def test_set_add(test_data, data):
for d in test_data:
data.add(d)
return data.__class__ .__name__ @cost_time
def test_list_add(test_data, data):
for d in test_data:
data.append(d)
return data.__class__ .__name__ #初始化数据,指定每种类型对应它添加元素的方法
def init_data():
test_data = {
'list': (list(), test_list_add),
'set': (set(), test_set_add),
'dict': (dict(), test_dict_add)
}
return test_data #每次检测10000增量大小的数据的添加时间
data_size = array([x for x in xrange(10000, 1000000, 10000)])
cost_result = {}
for size in data_size:
test_data = [x for x in xrange(size)]
for data_type, (data, add) in init_data().items():
name, cost = add(test_data, data) #返回方法的执行时间
cost_result.setdefault(data_type, []).append(cost) plot.figure(figsize=(10, 6))
xline = data_size
for data_type, result in cost_result.items():
yline = array(result)
plot.plot(xline, yline, label=data_type) plot.ylabel('Time spend')
plot.xlabel('Add times') plot.grid() plot.legend()
plot.show()

代码与图详解性能之Python集合类型(list tuple dict set generator)