python词频函数 python 词频分析
用Python统计词频
def statistics(astr):
在沙洋等地区,都构建了全面的区域性战略布局,加强发展的系统性、市场前瞻性、产品创新能力,以专注、极致的服务理念,为客户提供成都网站设计、网站建设 网站设计制作定制网站,公司网站建设,企业网站建设,品牌网站制作,成都全网营销推广,成都外贸网站制作,沙洋网站建设费用合理。
# astr.replace("\n", "")
slist = list(astr.split("\t"))
alist = []
[alist.append(i) for i in slist if i not in alist]
alist[-1] = alist[-1].replace("\n", "")
return alist
if __name__ == "__main__":
code_doc = {}
with open("test_data.txt", "r", encoding='utf-8') as fs:
for ln in fs.readlines():
l = statistics(ln)
for t in l:
if t not in code_doc:
code_doc.setdefault(t, 1)
else:
code_doc[t] += 1
for keys in code_doc.keys():
print(keys + ' ' + str(code_doc[keys]))
如何用python实现英文短文的双词频统计
简单版:
#!/usr/bin/env python3
import re
import jieba
from collections import Counter
fname = 'counttest.txt'
with open(fname) as f:
s = f.read()
pattern = re.compile(r'[a-zA-Z]+\-?[a-zA-Z]*')
english_words = Counter(pattern.findall(s))
other_words = Counter(jieba.cut(pattern.sub('', s)))
print('\n英文单词统计结果:\n'+'-'*17)
print('\n'.join(['{}: {}'.format(i, j) for i, j in english_words.most_common()]))
print('\n中文及符号统计结果:\n'+'-'*19)
print('\n'.join(['{}: {}'.format(i, j) for i, j in other_words.most_common()]))
复杂版:
#!/usr/bin/env python
# -*- coding: utf-8 -*-
from __future__ import print_function, division, unicode_literals
import sys, re, time, os, jieba
from collections import Counter
from datetime import datetime
class WordCounter(object):
def __init__(self, from_file, to_file=None, coding=None, jieba_cut=None):
'''根据设定的进程数,把文件from_file分割成大小基本相同,数量等同与进程数的文件段,
来读取并统计词频,然后把结果写入to_file中,当其为None时直接打印在终端或命令行上。
Args:
@from_file 要读取的文件
@to_file 结果要写入的文件
@coding 文件的编码方式,默认为采用chardet模块读取前1万个字符来自动判断
@jieba_cut 是否启用结巴分词,默认为None
How to use:
w = WordCounter('a.txt', 'b.txt')
w.run()
'''
if not os.path.isfile(from_file):
raise Exception('No such file: 文件不存在')
self.f1 = from_file
self.filesize = os.path.getsize(from_file)
self.f2 = to_file
if coding is None:
try:
import chardet
except ImportError:
os.system('pip install chardet')
print('-'*70)
import chardet
with open(from_file, 'rb') as f:
coding = chardet.detect(f.read(10000))['encoding']
self.coding = coding
self._c = [Counter(), Counter()]
self.jieba = False
if jieba_cut is not None:
self.jieba = True
def run(self):
start = time.time()
if 1:
self.count_direct(self.f1)
if self.f2 not in ['None', 'Null', 'none', 'null', None]:
with open(self.f2, 'wb') as f:
f.write(self.result.encode(self.coding))
else:
print('\nEnglish words:\n' + '-'*15)
print(self.result)
cost = '{:.1f}'.format(time.time()-start)
size = humansize(self.filesize)
tip = '\nFile size: {}. Cost time: {} seconds'
# print(tip.format(size, cost))
self.cost = cost + 's'
def count_direct(self, from_file):
'''直接把文件内容全部读进内存并统计词频'''
start = time.time()
with open(from_file, 'rb') as f:
line = f.read()
for i in range(len(self._c)):
self._c[i].update(self.parse(line)[i])
def parse(self, line): #解析读取的文件流
text = line.decode(self.coding)
text = re.sub(r'\-\n', '', text) #考虑同一个单词被分割成两段的情况,删除行末的-号
pattern = re.compile(r'[a-zA-Z]+\-?[a-zA-Z]*') #判断是否为英文单词
english_words = pattern.findall(text)
rest = pattern.sub('', text)
ex = Counter(jieba.cut(rest)) if self.jieba else Counter(text)
return Counter(english_words), ex
def flush(self): #清空统计结果
self._c = [Counter(), Counter()]
@property
def counter(self): #返回统计结果的Counter类
return self._c
@property
def result(self): #返回统计结果的字符串型式,等同于要写入结果文件的内容
ss = []
for c in self._c:
ss.append(['{}: {}'.format(i, j) for i, j in c.most_common()])
tip = '\n\n中文及符号统计结果:\n'+'-'*15+'\n'
return tip.join(['\n'.join(s) for s in ss])
def humansize(size):
"""将文件的大小转成带单位的形式
humansize(1024) == '1 KB'
True
humansize(1000) == '1000 B'
True
humansize(1024*1024) == '1 M'
True
humansize(1024*1024*1024*2) == '2 G'
True
"""
units = ['B', 'KB', 'M', 'G', 'T']
for unit in units:
if size 1024:
break
size = size // 1024
return '{} {}'.format(size, unit)
def main():
if len(sys.argv) 2:
print('Usage: python wordcounter.py from_file to_file')
exit(1)
from_file, to_file = sys.argv[1:3]
args = {'coding' : None, 'jieba_cut': 1}
for i in sys.argv:
for k in args:
if re.search(r'{}=(.+)'.format(k), i):
args[k] = re.findall(r'{}=(.+)'.format(k), i)[0]
w = WordCounter(from_file, to_file, **args)
w.run()
if __name__ == '__main__':
import doctest
doctest.testmod()
main()
更复杂的:如果是比较大的文件,建议采用多进程,详情百度:多进程读取大文件并统计词频 jaket5219999
如何用python对文章中文分词并统计词频
1、全局变量在函数中使用时需要加入global声明
2、获取网页内容存入文件时的编码为ascii进行正则匹配时需要decode为GB2312,当匹配到的中文写入文件时需要encode成GB2312写入文件。
3、中文字符匹配过滤正则表达式为ur'[\u4e00-\u9fa5]+',使用findall找到所有的中文字符存入分组
4、KEY,Value值可以使用dict存储,排序后可以使用list存储
5、字符串处理使用split分割,然后使用index截取字符串,判断哪些是名词和动词
6、命令行使用需要导入os,os.system(cmd)
如何用python和jieba分词,统计词频?
#! python3
# -*- coding: utf-8 -*-
import os, codecs
import jieba
from collections import Counter
def get_words(txt):
seg_list = jieba.cut(txt)
c = Counter()
for x in seg_list:
if len(x)1 and x != '\r\n':
c[x] += 1
print('常用词频度统计结果')
for (k,v) in c.most_common(100):
print('%s%s %s %d' % (' '*(5-len(k)), k, '*'*int(v/3), v))
if __name__ == '__main__':
with codecs.open('19d.txt', 'r', 'utf8') as f:
txt = f.read()
get_words(txt)
新闻名称:python词频函数 python 词频分析
文章出自:http://scyanting.com/article/dosgioj.html