This blog post is an introduction to regular expressions using Python.
All programming eventually boils down to operations on data, be it analysis, or data extraction, or transformation.
Before you can do anything with textual data, you have to first identify it in a very specific way. Your program, no matter what its going to do, must first be able to locate specific strings.
You may be trying to identify:
- whether a specific word is found in a file
- all email addresses referenced in a file
- all questions in a file (sentences ending with
- all content contained in
<p>tags in an html file
All of these cases describe some kind of text pattern.
Regular expressions are the definition of such patterns which allow you to lock onto any textual data that matches the pattern.
Regexes are notorious for being arcane and difficult to read. Even simple expressions can look like gibberish at first glance, which is why so many avoid using them altogether. I used to avoid them like the plague when I first learned about them, largely because there wasn’t a tool to give me real-time feedback on my regex.
Those days are long gone, with the introduction of the pythex.org website!
This site is a great tool to help build and test regular expressions, because you can reference a cheat sheet and get immediate results on what is being matched.
Here are some basic matching rules we will be using in the demo:
.matches any character
\smatches any whitespace in the set
[ \t\n\r\f\v ]
\wmatches alphanumeric characters in the set
@matches the literal
\.matches the literal period
There are certain grouping constructs available in regular expressions:
( )specify your capture group by enclosing a pattern. Your captures are whatever strings end up matching the pattern enclosed.
- Square brackets
[ ]match any characters enclosed, with no regard to sequence.
With this knowledge, we can build a basic pattern to match email addresses contained in a string.
The test string is:
Here are some emails. They can be strewn about in this string in any haphazard way. Like email@example.com Or firstname.lastname@example.org. Followed by ronWeasly@hogwarts.edu
The information we want to capture is something@something followed by .com or .edu. This is our pattern which we can now start building formally.
To match any alphanumeric characters before and after the
@ symbol, we can start with the
\w matcher, all enclosed in parentheses to indicate our capture group.
Trying that in pythex, you’ll see that it’s not exactly what we want.
The captured values are:
e@o r@h y@h
We need some way of saying one or more characters.
These are provided to us as quantifiers:
*matches 0 or more
+matches 1 or more
?matches 0 or 1
There are more quantifiers that let us specify exact, minimum or maximum occurrances.
If we change our regex to
(\w+@\w+), we see some progress!
doe@outlook potter@hogwarts ronWeasly@hogwarts
The period character is causing the first names to be left out, since
\w only matches numbers, alphabets and underscores. We want one or more alphanumeric characters or periods in any sequence.
Since we have an
or, we can now start using brackets
[ ], the character set matcher.
The period already has a special meaning, which is to match any character, so we need to escape it with a backslash prefix. Our match set looks like
[\w\.] which reads any alphanuremic character or period. The quantity is still singular.
If we add the
+ quantifier at the end, then
[\w\.]+ reads one or more alphanumeric characters or periods in any sequence. This gets us much closer.
john.doe@outlook harry.potter@hogwarts ronWeasly@hogwarts
To handle the
.edu part, we can specify another
or set but using brackets doesn’t work.
Brackets are for matching any singular character in the set, not an ordered sequence of characters.
In this case, we need a grouping mechanism that will let us match against optional ordered sequences of characters.
We can enclose our .com and .edu inside
(?: ) which is the non-capturing version of the regular parentheses.
(?:\.com|\.edu) reads either .com or .edu. The pipe
| symbol stands for
or. Since the
\. part is common to both, we can place it before the group.
Our regex now looks like
([\.\w]+@\w+\.(?:com|edu)) and captures:
email@example.com firstname.lastname@example.org ronWeasly@hogwarts.edu
If we left out the
?: part above, an interesting thing happens. We then have 2 capturing groups as seen here.
This reveals another cool aspect of regular expressions which is the ability to capture elements within a matched substring. By using
(?:com|edu), we are saying presence of com or edu without actually storing the
edu text as a captured results.
If we leave it out, then
edu become distinct matched elements themselves. It doesn’t affect our output that much because we still have the complete emails captured, but the extra captures don’t look clean.
By modifying the last part a little bit, we can make it more generic and able to handle any domain ending sequence like
We can describe the ending part as an ordered sequence of a period
\. followed by any word
A little shorter, a little more clearer. The completed example is available here.
The next example introduces another useful regex matcher.
Suppose we have an HTML file and want to extract all the content in
<html> <head> <title> A Simple HTML Document </title> </head> <body> <p>This is a very simple HTML document</p> <p>It only has two paragraphs</p> </body> </html>
In a real world scenario, it might be a page hundreds of lines long, and doing so manually would be immensely time consuming. Without regular expressions, your algorithm might be to read each line of the html file, check to see if there is a
<p> string in the line, and start taking slices of each line until reaching a
A regular expression can really help us here. Just like last time, we’ll first try to describe our pattern in plain english.
We want to capture all characters preceded by
<p> and succeeded by
(?<= ) group describes a look behind assertion.
For example, in
the quick brown fox jumps over the lazy dog, the regex
(?<=the)\s(\w+) will match
(?<= ) reads preceded by whatever is enclosed. The
\s character denotes any singular whitespace, and is
not part of our capture group
These look behind assertions are very useful whenever you know the prefix of your capture group. For our HTML tag, it would be
We want everything between the
</p> tags. May I present the all capturer,
If you just run this regex on anything, you’ll match the entire string.
The closing tag can be expressed with the non-capturing group
Here is a link to the completed expression, including our html test string.
To use regular expressions in Python, we import the
re module from the standard library.
The documentation is very heplful and can be found here.
import re def main(): with open("test.html", "rt") as f: html = f.read() paragrphs = get_paragraphs(html) [print(p + "\n") for p in paragrphs] def get_paragraphs(html): return re.findall(r"(?<=<p>)(.*)(?:</p>)", html) if __name__ == "__main__": main()
Running this script gives the following output, when the html is stored in
test.html in the same directory:
> python paragraphs.py This is a very simple HTML document It only has two paragraphs
Witness the power of regular expressions!
The task of identifying my data boiled down to this function call which does all the heavy lifting:
r prefixed string is a raw string which treats the backslash
\ character as a literal character instead of a special character. For example,
r"\n" is literally a backslash followed by an n, not a newline character as a whole. It’s best practice to write regular expressions as raw strings.
While these examples are not very complicated, I hope they helped touch on some great functionality provided by regular expressions.
I use them in my work almost every day, and now you can start using them too!