Read txt file with multi-threaded in python

Question

I'm trying to read a file in python (scan it lines and look for terms) and write the results- let say, counters for each term. I need to do that for a big amount of files (more than 3000). Is it possible to do that multi threaded? If yes, how?

So, the scenario is like this:

• Read each file and scan its lines
• Write counters to same output file for all the files I've read.

Second question is, does it improve the speed of read/write.

Hope it is clear enough. Thanks,

Ron.

Answer 1

I agree with @aix, multiprocessing is definitely the way to go. Regardless you will be i/o bound -- you can only read so fast, no matter how many parallel processes you have running. But there can easily be some speedup.

Consider the following (input/ is a directory that contains several .txt files from Project Gutenberg).

import os.path
from multiprocessing import Pool
import sys
import time

def process_file(name):
    ''' Process one file: count number of lines and words '''
    linecount=0
    wordcount=0
    with open(name, 'r') as inp:
        for line in inp:
            linecount+=1
            wordcount+=len(line.split(' '))

    return name, linecount, wordcount

def process_files_parallel(arg, dirname, names):
    ''' Process each file in parallel via Poll.map() '''
    pool=Pool()
    results=pool.map(process_file, [os.path.join(dirname, name) for name in names])

def process_files(arg, dirname, names):
    ''' Process each file in via map() '''
    results=map(process_file, [os.path.join(dirname, name) for name in names])

if __name__ == '__main__':
    start=time.time()
    os.path.walk('input/', process_files, None)
    print "process_files()", time.time()-start

    start=time.time()
    os.path.walk('input/', process_files_parallel, None)
    print "process_files_parallel()", time.time()-start

When I run this on my dual core machine there is a noticeable (but not 2x) speedup:

$ python process_files.py
process_files() 1.71218085289
process_files_parallel() 1.28905105591

If the files are small enough to fit in memory, and you have lots of processing to be done that isn't i/o bound, then you should see even better improvement.

Answer 2

Yes, it should be possible to do this in a parallel manner.

However, in Python it's hard to achieve parallelism with multiple threads. For this reason multiprocessing is the better default choice for doing things in parallel.

It is hard to say what kind of speedup you can expect to achieve. It depends on what fraction of the workload it will be possible to do in parallel (the more the better), and what fraction will have to be done serially (the less the better).