timeit与时间装饰器

时间:2009-10-26 03:23:34

标签: python timing timeit

我正在尝试计算一些代码。首先,我使用了时间装饰器:

#!/usr/bin/env python

import time
from itertools import izip
from random import shuffle

def timing_val(func):
    def wrapper(*arg, **kw):
        '''source: http://www.daniweb.com/code/snippet368.html'''
        t1 = time.time()
        res = func(*arg, **kw)
        t2 = time.time()
        return (t2 - t1), res, func.__name__
    return wrapper

@timing_val
def time_izip(alist, n):
    i = iter(alist)
    return [x for x in izip(*[i] * n)]

@timing_val
def time_indexing(alist, n):
    return [alist[i:i + n] for i in range(0, len(alist), n)]

func_list = [locals()[key] for key in locals().keys()
             if callable(locals()[key]) and key.startswith('time')]
shuffle(func_list)  # Shuffle, just in case the order matters

alist = range(1000000)
times = []
for f in func_list:
    times.append(f(alist, 31))

times.sort(key=lambda x: x[0])
for (time, result, func_name) in times:
    print '%s took %0.3fms.' % (func_name, time * 1000.)

产量

% test.py
time_indexing took 73.230ms.
time_izip took 122.057ms.

在这里我使用timeit:

%  python - m timeit - s '' 'alist=range(1000000);[alist[i:i+31] for i in range(0, len(alist), 31)]'
10 loops, best of 3:
    64 msec per loop
% python - m timeit - s 'from itertools import izip' 'alist=range(1000000);i=iter(alist);[x for x in izip(*[i]*31)]'
10 loops, best of 3:
    66.5 msec per loop

使用timeit结果几乎相同,但使用时序装饰器时,time_indexing的显示速度比time_izip快。

这种差异的原因是什么?

是否应该相信哪种方法?

若然,哪个?

8 个答案:

答案 0 :(得分:49)

使用functools的包裹来改善Matt Alcock的答案。

from functools import wraps
from time import time

def timing(f):
    @wraps(f)
    def wrap(*args, **kw):
        ts = time()
        result = f(*args, **kw)
        te = time()
        print 'func:%r args:[%r, %r] took: %2.4f sec' % \
          (f.__name__, args, kw, te-ts)
        return result
    return wrap

在一个例子中:

@timing
def f(a):
    for _ in range(a):
        i = 0
    return -1

调用f包裹的方法@timing

func:'f' args:[(100000000,), {}] took: 14.2240 sec
f(100000000)

这样做的好处是它保留了原始功能的属性;也就是说,函数名称和docstring等元数据在返回的函数上被正确保留。

答案 1 :(得分:29)

我会使用一个时序装饰器,因为你可以使用注释来围绕你的代码撒上时序,而不是让你的代码混乱时序逻辑。

import time

def timeit(f):

    def timed(*args, **kw):

        ts = time.time()
        result = f(*args, **kw)
        te = time.time()

        print 'func:%r args:[%r, %r] took: %2.4f sec' % \
          (f.__name__, args, kw, te-ts)
        return result

    return timed

使用装饰器很容易使用注释。

@timeit
def compute_magic(n):
     #function definition
     #....

或者重新设置您想要的时间功能。

compute_magic = timeit(compute_magic)

答案 2 :(得分:18)

使用timeit。不止一次地运行测试会给我带来更好的结果。

func_list=[locals()[key] for key in locals().keys() 
           if callable(locals()[key]) and key.startswith('time')]

alist=range(1000000)
times=[]
for f in func_list:
    n = 10
    times.append( min(  t for t,_,_ in (f(alist,31) for i in range(n)))) 

for (time,func_name) in zip(times, func_list):
    print '%s took %0.3fms.' % (func_name, time*1000.)

- >

<function wrapper at 0x01FCB5F0> took 39.000ms.
<function wrapper at 0x01FCB670> took 41.000ms.

答案 3 :(得分:6)

我厌倦了from __main__ import foo,现在使用它 - 对于简单的args,%r工作, 而不是在Ipython中 (为什么timeit仅适用于字符串,而不是thunks / closures,即timefunc(f,任意args)?)


import timeit

def timef( funcname, *args, **kwargs ):
    """ timeit a func with args, e.g.
            for window in ( 3, 31, 63, 127, 255 ):
                timef( "filter", window, 0 )
    This doesn't work in ipython;
    see Martelli, "ipython plays weird tricks with __main__" in Stackoverflow        
    """
    argstr = ", ".join([ "%r" % a for a in args]) if args  else ""
    kwargstr = ", ".join([ "%s=%r" % (k,v) for k,v in kwargs.items()]) \
        if kwargs  else ""
    comma = ", " if (argstr and kwargstr)  else ""
    fargs = "%s(%s%s%s)" % (funcname, argstr, comma, kwargstr)
        # print "test timef:", fargs
    t = timeit.Timer( fargs, "from __main__ import %s" % funcname )
    ntime = 3
    print "%.0f usec %s" % (t.timeit( ntime ) * 1e6 / ntime, fargs)

#...............................................................................
if __name__ == "__main__":
    def f( *args, **kwargs ):
        pass

    try:
        from __main__ import f
    except:
        print "ipython plays weird tricks with __main__, timef won't work"
    timef( "f")
    timef( "f", 1 )
    timef( "f", """ a b """ )
    timef( "f", 1, 2 )
    timef( "f", x=3 )
    timef( "f", x=3 )
    timef( "f", 1, 2, x=3, y=4 )

补充:另请参阅“ipython与主要一起玩奇怪的技巧”,Martelli 在running-doctests-through-ipython

答案 4 :(得分:2)

只是一个猜测,但差异可能是范围()值差异的数量级?

来自您的原始来源:

alist=range(1000000)

来自timeit示例:

alist=range(100000)

对于它的价值,以下是我的系统上的结果,范围设置为100万:

$ python -V
Python 2.6.4rc2

$ python -m timeit -s 'from itertools import izip' 'alist=range(1000000);i=iter(alist);[x for x in izip(*[i]*31)]'
10 loops, best of 3: 69.6 msec per loop

$ python -m timeit -s '' 'alist=range(1000000);[alist[i:i+31] for i in range(0, len(alist), 31)]'
10 loops, best of 3: 67.6 msec per loop

我无法运行其他代码,因为我无法在我的系统上导入“装饰器”模块。


更新 - 我发现在没有装饰器的情况下运行代码时会出现同样的差异。

$ ./test.py
time_indexing took 84.846ms.
time_izip took 132.574ms.

感谢您发布此问题;我今天学到了一些东西=)

答案 5 :(得分:2)

这是您祈祷图书馆提供一种便携式解决方案的需求-DRY!幸运的是,funcy.log_durations来了。

示例摘自文档:

@log_durations(logging.info)
def do_hard_work(n):
    samples = range(n)
    # ...

# 121 ms in do_hard_work(10)
# 143 ms in do_hard_work(11)
# ...

浏览功能文档以了解其他变体,例如不同的关键字参数和@log_iter_durations

答案 6 :(得分:1)

受米卡·史密斯(Micah Smith)的回答启发,我直接进行了有趣的打印(而不使用日志记录模块)。

以下在Google colab上方便使用。

# pip install funcy
from funcy import print_durations

@print_durations()
def myfunc(n=0):
  for i in range(n):
    pass

myfunc(123)
myfunc(123456789)

# 5.48 mks in myfunc(123)
# 3.37 s in myfunc(123456789)

答案 7 :(得分:0)

无论这个特定的练习如何,我都认为使用timeit是更安全可靠的选择。与您的解决方案不同,它也是跨平台的。