需要更快的地图功能

Question

我想优化我的代码以在hadoop集群上运行。任何人都可以帮助我找到一些方法来改善这一点吗？我正在接受一组数量超过四千万的数字，每个数字都在一个新的线上。随着数字读入我计算每个数字，总结所有数字，并检查每个数字，看看它是否是一个素数。

#!/usr/bin/env python 

import sys 
import string 
import math 

total_of_primes = 0 
total = 0 
count = 0 
not_prime = 0 
count_string = 'Count:' 
total_string = 'Total:' 
prime_string = 'Number of Primes:' 

for line in sys.stdin: 
    try: 
    key = int(line) 
    except: 
    continue 
    total = total + key 
    count = count + 1 
    if key == 2 or key == 3: 
    not_prime = not_prime - 1 
    elif key%2 == 0 or key%3 == 0: 
    not_prime = not_prime + 1 
    else: 
    for i in range(5,(int(math.sqrt(key))+1),6): 
     if key%i == 0 or key%(i+2) ==0: 
     not_prime = not_prime + 1 
     break 

total_of_primes = count - not_prime 


print '%s\t%s' % (count_string,count) 
print '%s\t%s' % (total_string,total) 
print '%s\t%s' % (prime_string,total_of_primes) 

Answer 1

我试着把一切都变成理解。理解比原生Python代码更快，因为他们访问C库。我也省略了2和3的测试，因为您可以在完成循环后手动添加这些测试。

我几乎可以保证这会有错误，因为我没有你的测试数据，并且对于我来说（对我来说）这个大的理解真的需要测试。这在技术上是一个单线，但我试图将其拆分为可读性。不过，希望至少能给你一些想法。

biglist = [ # this will be a list of booleans 
    not int(line)%2 or # the number is not even 
    not int(line)%3 or # not divisible by 3 
    (
     not int(line)%i or # not divisible by each item in the range() object 
     not int(line)%(i+2) for i in # nor by 2 greater than each item 
      # and only go through the range() object while it's still prime 
      itertools.takewhile(lambda x: not int(line)%x or not int(line)%(x+2), 
     range(5, int(pow(int(line), 0.5))+1, 6)) # pow(x, 0.5) uses a built-in instead of an imported module 
    ) 
for line in sys.stdin) if line.lstrip('-+').isdigit() # going through each item in sys.stdin 
# as long as long as it's a digit. if you only expect positive numbers, you can omit ".lstrip('-+')". 
] 

total_of_primes = len(biglist) + 2 # manually add 2 and 3 instead of testing it 

如果你不能得到执行时间下来远远不够，你可能会考虑迁移到较低级别（慢写，运行更快）的语言，像C.