需要更快的地图功能

Question

我正在尝试优化我的代码以在 hadoop 集群上运行。谁能帮我找到一些方法来改善它？我正在接受一组非常大的数字 40+ 百万，每个数字都在一个新的行上。在读入数字时，我正在计算每个数字，将所有数字相加，并检查每个数字是否为素数。

#!/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

我试图把一切都变成一种理解。理解比原生 Python 代码更快，因为它们访问 C 库。我也省略了对2 和3 的测试，因为您可以在完成循环后手动添加它们。

我几乎可以保证这会有错误，因为我没有你的测试数据和这么大的理解（无论如何对我来说）确实需要测试。从技术上讲，它是单行的，但为了便于阅读，我尝试将其拆分。不过，希望它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

如果您无法将执行时间缩短到足够长，您可能会考虑转向较低级别（编写速度较慢，运行速度较快）的语言，例如 C。