python自然语言处理：有效信息的提取

最近跑了3D CNN，实现MRI的分割及分类。训练过程中将准确率等自动保存成一个txt文件，该文件有64M多，想要从中提取有用信息比如Mean accuracy、sensitivity、specificity和mean dsc等，人工筛选太繁琐，所以想到用python编程来实现。

去请教了一位研究自然语言处理的计算机系的师兄，实现代码如下：

	# -*- coding: utf-8 -*-
	# @Time : 18-4-18
	# @Author : sadscv
	# @File : textExtract.py
	import json
	import re
	def readfile():
	save = None
	count = 0
	with open("data/trainSessionDeepMedic.txt", "r") as f:
	tmpfile = open("data/Epoch.txt", "wb")
	# split file
	for line in f:
	print(save)
	head = re.search("Starting new Epoch! Epoch", line, flags=0)
	if head:
	pos = re.search("#", line).span()[1]
	num = line[pos:pos + 2].rstrip("/")
	save = num
	open("data/Epoch_{}.txt".format(save), "wb")
	count += 1
	tmpfile = open("data/Epoch_{}.txt".format(save), "a")
	elif save: