一、什么是正则表达式

正则表达式是一种特殊的字符串模式，用于匹配一组字符串，就好比用模具做产品，而正则就是这个模具，定义一种规则去匹配符合规则的字符。

1.2 常用的正则匹配工具

在线匹配工具：
1. http://www.regexpal.com/
2. http://rubular.com/
3. http://tools.jb51.net/regex/create_reg
4. txt2re ：这个在线网站支持解析一句话，从中可以生成匹配的正则表达式，且可以生成诸多类型的代码。语言支持：Perl PHP Python Java Javascript ColdFusion C C++ Ruby VB VBScript J#.net C#.net C++.net VB.net

正则匹配软件

McTracer

用过几个之后还是觉得这个是最好用的，支持将正则导成对应的语言如java C# js等还帮你转义了，Copy直接用就行了很方便，另外支持把正则表达式用法解释，如哪一段是捕获分组，哪段是贪婪匹配等等，总之用起来 So Happy .

二正则字符简单介绍

关于这部分建议跳到：《读懂正则表达式就这么简单》
里面有很详细的介绍。

另外关于python的正则表达式，主要使用re模块。

我们以任务为导向介绍python正则表达式的用法。
假设给我们下面这段话：

I1113 23:35:50.763059  4460 solver.cpp:218] Iteration 400 (27.3075 iter/s, 0.7324s/20 iters), loss = 0.0202583
I1113 23:35:50.763141  4460 solver.cpp:237]     Train net output #0: rpn_cls_loss = 0.00101873 (* 1 = 0.00101873 loss)
I1113 23:35:50.763165  4460 solver.cpp:237]     Train net output #1: rpn_loss_bbox = 0.0192396 (* 1 = 0.0192396 loss)
I1113 23:35:50.763175  4460 sgd_solver.cpp:105] Iteration 400, lr = 0.001
I1113 23:35:51.751206  4460 solver.cpp:218] Iteration 420 (20.2456 iter/s, 0.987868s/20 iters), loss = 0.00228514
I1113 23:35:51.751341  4460 solver.cpp:237]     Train net output #0: rpn_cls_loss = 0.00140554 (* 1 = 0.00140554 loss)
I1113 23:35:51.751379  4460 solver.cpp:237]     Train net output #1: rpn_loss_bbox = 0.000879596 (* 1 = 0.000879596 loss)
I1113 23:35:51.751410  4460 sgd_solver.cpp:105] Iteration 420, lr = 0.001
I1113 23:35:52.523890  4460 solver.cpp:218] Iteration 440 (25.8933 iter/s, 0.772401s/20 iters), loss = 0.0132958
I1113 23:35:52.523974  4460 solver.cpp:237]     Train net output #0: rpn_cls_loss = 0.00312161 (* 1 = 0.00312161 loss)
I1113 23:35:52.523988  4460 solver.cpp:237]     Train net output #1: rpn_loss_bbox = 0.0101742 (* 1 = 0.0101742 loss)
I1113 23:35:52.523998  4460 sgd_solver.cpp:105] Iteration 440, lr = 0.001
I1113 23:35:53.461998  4460 solver.cpp:218] Iteration 460 (21.3325 iter/s, 0.937539s/20 iters), loss = 0.0154897
I1113 23:35:53.462057  4460 solver.cpp:237]     Train net output #0: rpn_cls_loss = 0.00780452 (* 1 = 0.00780452 loss)
I1113 23:35:53.462069  4460 solver.cpp:237]     Train net output #1: rpn_loss_bbox = 0.00768522 (* 1 = 0.00768522 loss)
I1113 23:35:53.462082  4460 sgd_solver.cpp:105] Iteration 460, lr = 0.001
I1113 23:35:54.356657  4460 solver.cpp:218] Iteration 480 (22.3584 iter/s, 0.894517s/20 iters), loss = 0.00275768
I1113 23:35:54.356729  4460 solver.cpp:237]     Train net output #0: rpn_cls_loss = 0.00107937 (* 1 = 0.00107937 loss)
I1113 23:35:54.356739  4460 solver.cpp:237]     Train net output #1: rpn_loss_bbox = 0.00167831 (* 1 = 0.00167831 loss)
I1113 23:35:54.356748  4460 sgd_solver.cpp:105] Iteration 480, lr = 0.001
I1113 23:35:55.153437  4460 solver.cpp:218] Iteration 500 (25.1734 iter/s, 0.79449s/20 iters), loss = 0.0230187
I1113 23:35:55.153519  4460 solver.cpp:237]     Train net output #0: rpn_cls_loss = 0.0105348 (* 1 = 0.0105348 loss)
I1113 23:35:55.153530  4460 solver.cpp:237]     Train net output #1: rpn_loss_bbox = 0.0124839 (* 1 = 0.0124839 loss)
I1113 23:35:55.153542  4460 sgd_solver.cpp:105] Iteration 500, lr = 0.001
I1113 23:35:56.104395  4460 solver.cpp:218] Iteration 520 (21.0352 iter/s, 0.950785s/20 iters), loss = 0.0144106
I1113 23:35:56.104485  4460 solver.cpp:237]     Train net output #0: rpn_cls_loss = 0.00135394 (* 1 = 0.00135394 loss)
I1113 23:35:56.104504  4460 solver.cpp:237]     Train net output #1: rpn_loss_bbox = 0.0130567 (* 1 = 0.0130567 loss)
I1113 23:35:56.104521  4460 sgd_solver.cpp:105] Iteration 520, lr = 0.001
I1113 23:35:56.854631  4460 solver.cpp:218] Iteration 540 (26.6699 iter/s, 0.749909s/20 iters), loss = 0.0167331
I1113 23:35:56.854696  4460 solver.cpp:237]     Train net output #0: rpn_cls_loss = 0.00285695 (* 1 = 0.00285695 loss)
I1113 23:35:56.854710  4460 solver.cpp:237]     Train net output #1: rpn_loss_bbox = 0.0138762 (* 1 = 0.0138762 loss)
I1113 23:35:56.854720  4460 sgd_solver.cpp:105] Iteration 540, lr = 0.001
I1113 23:35:57.824692  4460 solver.cpp:218] Iteration 560 (20.6206 iter/s, 0.969902s/20 iters), loss = 0.00817935
I1113 23:35:57.824774  4460 solver.cpp:237]     Train net output #0: rpn_cls_loss = 0.00557839 (* 1 = 0.00557839 loss)
I1113 23:35:57.824791  4460 solver.cpp:237]     Train net output #1: rpn_loss_bbox = 0.00260096 (* 1 = 0.00260096 loss)
I1113 23:35:57.824806  4460 sgd_solver.cpp:105] Iteration 560, lr = 0.001
I1113 23:35:58.670575  4460 solver.cpp:218] Iteration 580 (23.6486 iter/s, 0.845714s/20 iters), loss = 0.00420315
I1113 23:35:58.670637  4460 solver.cpp:237]     Train net output #0: rpn_cls_loss = 0.0020043 (* 1 = 0.0020043 loss)
I1113 23:35:58.670648  4460 solver.cpp:237]     Train net output #1: rpn_loss_bbox = 0.00219884 (* 1 = 0.00219884 loss)
I1113 23:35:58.670658  4460 sgd_solver.cpp:105] Iteration 580, lr = 0.001
I1114 00:34:17.348683  4460 sgd_solver.cpp:105] Iteration 79980, lr = 0.0001
speed: 0.044s / iter
Wrote snapshot to: /data1/caiyong.wang/program/py-faster-rcnn/output/faster_rcnn_alt_opt/voc_2007_trainval/zf_rpn_stage1_iter_80000.caffemodel

希望我们解析出

Iteration 500 (25.1734 iter/s, 0.79449s/20 iters), loss = 0.0230187

中的Iteration与loss值。 其实这是faster rcnn生成的log文件一部分。

我们通过上面的语法学习，在MTracer中生成了正则表达式：

\bIteration\s(?<Iteration>\d+)\s\(.*\).*loss\s=\s(?<loss>\d*\.*\d+)\b

注：.*表示除换行符以外的任意字符，*表示0个或多个
并且采用多行模式解析出了我们想要的结果。
上面的表达式中，我们使用了捕获分组。
如下图：

那么如何转化成python代码？
正确的代码如下：

import repattern = re.compile(r'\bIteration\s(?P<Iteration>\d+)\s\(.*\).*loss\s=\s(?P<loss>\d*\.*\d+)\b')
arr=pattern.search("I1113 23:35:50.763059  4460 solver.cpp:218] Iteration 400 (27.3075 iter/s, 0.7324s/20 iters), loss = 0.0202583")
arr.groups()
arr.group()
arr.group("Iteration")
arr.group("loss")

结果为：

arr.groups()
Out[147]: ('400', '0.0202583')arr.group()
Out[148]: 'Iteration 400 (27.3075 iter/s, 0.7324s/20 iters), loss = 0.0202583'arr.group("Iteration")
Out[149]: '400'arr.group("loss")
Out[150]: '0.0202583'

这里python的命名组与以往的不一样，使用的是 (?P<name>exp) 取代(?<name>exp)

而且compile里面必须加上r。

参考文献：

http://blog.csdn.net/lwnylslwnyls/article/details/8901273
https://www.cnblogs.com/tk091/p/3702307.html
PYTHON的RE模块理解（RE.COMPILE、RE.MATCH、RE.SEARCH）

三 python正则表达式的其他用法。

多行匹配：

python多行匹配

r = re.compile("需要的正则表达式", re.M)

匹配到需要的字符，可以获取红括号内的数字

r = re.compile("r([0-9]{5,})")

举个例子：需要获取20462和24729连个数字

import re
data = """r24062 line1hello word !!!!r24729 line2revision:24181"""r = re.compile("^r([0-9]{5,})", re.M)nums = r.findall(data)print nums---------------------
output:["24062", "24729"]

注：{5,}表示至少重复5次
2. 无捕获分组
当你要将一部分规则作为一个整体对它进行某些操作，比如指定其重复次数时，你需要将这部分规则用