We Keep Coding

Python regex: re.search() is extremely slow on large text files

My code does the following:
Take a large text file (i.e. a legal document that is 300 pages as a PDF).
Find a certain keyword (e.g. "small").
Return n words to the left and n words to the right of the keyword.
NOTE: In this context, a "word" is any string of non-space characters. "$cow123" would be a word, but "health care" would be two words.
Here is my problem:
The code takes an extremely long time to run on the 300 pages, and that time tends to increase very quickly as n increases.
Here is my code:
fileHandle = open('test_pdf.txt', mode='r')
document = fileHandle.read()
def search(searchText, doc, n):
#Searches for text, and retrieves n words either side of the text, which are returned separately
surround = r"\s*(\S*)\s*"
groups = re.search(r'{}{}{}'.format(surround*n, searchText, surround*n), doc).groups()
return groups[:n],groups[n:]
Here is the nasty culprit:
print search("\$27.5 million", document, 10)
Here's how you can test this code:
Copy the function definition from the code block above and run the following:
t = "The world is a small place, we $.205% try to take care of it."
print search("\$.205", t, 3)
I suspect that I have a nasty case of catastrophic backtracking, but I'm too new to regex to point my finger on the problem.
How do I speed up my code?

How about using re.search (or even string.find if you're only searching for fixed strings) to find the string, without any surrounding capturing groups. Then you use the position and length of the match (.start and .end on a re matchobject, or the return value of find plus the length of the search string). Get the substring before the match and do /\s*(\S*)\s*\z/ etc. on it, and get the substring after the match and do /\A\s*(\S*)\s*/ etc. on it.
Also, for help with your backtracking: you can use a pattern like \s+\S+\s+ instead of \s*\S*\s* (two chunks of whitespace have to be separated by a non-zero amount of non-whitespace, or else they wouldn't be two chunks), and you shouldn't butt up two consecutive \s*s like you do. I think r'\S+'.join([[r'\s+']*(n)) would give the right pattern for capturing n previous words (but my Python is rusty, so check that).

I see several problems here. The First, and probably worst, is that everything in your "surround" regex is, not just optional but independently optional. Given this string:
"Lorem ipsum tritani impedit civibus ei pri"
...when searchText = "tritani" and n = 1, this is what it has to go through before it finds the first match:
regex: \s* \S* \s* tritani
offset 0: '' 'Lorem' ' ' FAIL
'' 'Lorem' '' FAIL
'' 'Lore' '' FAIL
'' 'Lor' '' FAIL
'' 'Lo' '' FAIL
'' 'L' '' FAIL
'' '' '' FAIL
...then it bumps ahead one position and starts over:
offset 1: '' 'orem' ' ' FAIL
'' 'orem' '' FAIL
'' 'ore' '' FAIL
'' 'or' '' FAIL
'' 'o' '' FAIL
'' '' '' FAIL
... and so on. According to RegexBuddy's debugger, it takes almost 150 steps to reach the offset where it can make the first match:
position 5: ' ' 'ipsum' ' ' 'tritani'
And that's with just one word to skip over, and with n=1. If you set n=2 you end up with this:
\s*(\S*)\s*\s*(\S*)\s*tritani\s*(\S*)\s*\s*(\S*)\s*
I sure you can see where this is is going. Note especially that when I change it to this:
(?:\s+)(\S+)(?:\s+)(\S+)(?:\s+)tritani(?:\s+)(\S+)(?:\s+)(\S+)(?:\s+)
...it finds the first match in a little over 20 steps. This is one of the most common regex anti-patterns: using * when you should be using +. In other words, if it's not optional, don't treat it as optional.
Finally, you may have noticed the \s*\s* the auto-generated regex

You could try using mmap and appropriate regex flags, eg (untested):
import re
import mmap
with open('your file') as fin:
mf = mmap.mmap(fin.fileno(), 0, access=mmap.ACCESS_READ)
for match in re.finditer(your_re, mf, flags=re.DOTALL):
print match.group() # do something with your match
This'll only keep memory usage lower though...
The alternative is to have a sliding window of words (simple example of just single word before and after)...:
import re
import mmap
from itertools import islice, tee, izip_longest
with open('testingdata.txt') as fin:
mf = mmap.mmap(fin.fileno(), 0, access=mmap.ACCESS_READ)
words = (m.group() for m in re.finditer('\w+', mf, flags=re.DOTALL))
grouped = [islice(el, idx, None) for idx, el in enumerate(tee(words, 3))]
for group in izip_longest(*grouped, fillvalue=''):
if group[1] == 'something': # check criteria for group
print group

I think you are going about this completely backwards (I'm a little confused as to what you are doing in the first place!)
I would recommend checking out the re_search function I developed in the textools module of my cloud toolbox
with re_search you could solve this problem with something like:
from cloudtb import textools
data_list = textools.re_search('my match', pdf_text_str) # search for character objects
# you now have a list of strings and RegPart objects. Parse through them:
for i, regpart in enumerate(data_list):
if isinstance(regpart, basestring):
words = textools.re_search('\w+', regpart)
# do stuff with words
else:
# I Think you are ignoring these? Not totally sure
Here is a link on how to use and how it works:
http://cloudformdesign.com/?p=183
In addition to this, your regular expressions would also be printed out in more readable format.
You might also want to check out my tool Search The Sky or the similar tool Kiki to help you build and understand your regular expressions.

Python Regular Expression - right-to-left

I am trying to use regular expressions in python to match the frame number component of an image file in a sequence of images. I want to come up with a solution that covers a number of different naming conventions. If I put it into words I am trying to match the last instance of one or more numbers between two dots (eg .0100.). Below is an example of how my current logic falls down:
import os
import re
def sub_frame_number_for_frame_token(path, token='#'):
folder = os.path.dirname(path)
name = os.path.basename(path)
pattern = r'\.(\d+)\.'
matches = list(re.finditer(pattern, name) or [])
if not matches:
return path
# Get last match.
match = matches[-1]
frame_token = token * len(match.group(1))
start, end = match.span()
apetail_name = '%s.%s.%s' % (name[:start], frame_token, name[end:])
return os.path.join(folder, apetail_name)
# Success
eg1 = 'xx01_010_animation.0100.exr'
eg1 = sub_frame_number_for_frame_token(eg1) # result: xx01_010_animation.####.exr
# Failure
eg2 = 'xx01_010_animation.123.0100.exr'
eg2 = sub_frame_number_for_frame_token(eg2) # result: xx01_010_animation.###.0100.exr
I realise there are other ways in which I can solve this issue (I have already implemented a solution where I am splitting the path at the dot and taking the last item which is a number) but I am taking this opportunity to learn something about regular expressions. It appears the regular expression creates the groups from left-to-right and cannot use characters in the pattern more than once. Firstly is there anyway to search the string from right-to-left? Secondly, why doesn't the pattern find two matches in eg2 (123 and 0100)?
Cheers

finditer will return an iterator "over all non-overlapping matches in the string".
In your example, the last . of the first match will "consume" the first . of the second. Basically, after making the first match, the remaining string of your eg2 example is 0100.exr, which doesn't match.
To avoid this, you can use a lookahead assertion (?=), which doesn't consume the first match:
>>> pattern = re.compile(r'\.(\d+)(?=\.)')
>>> pattern.findall(eg1)
['0100']
>>> pattern.findall(eg2)
['123', '0100']
>>> eg3 = 'xx01_010_animation.123.0100.500.9000.1234.exr'
>>> pattern.findall(eg3)
['123', '0100', '500', '9000', '1234']
# and "right to left"
>>> pattern.findall(eg3)[::-1]
['1234', '9000', '500', '0100', '123']

My solution uses a very simple hackish way of fixing it. It reverses the string path in the beginning of your function and reverses the return value at the end of it. It basically uses regular expressions to search the backwards version of your given strings. Hackish, but it works. I used the syntax shown in this question to reverse the string.
import os
import re
def sub_frame_number_for_frame_token(path, token='#'):
path = path[::-1]
folder = os.path.dirname(path)
name = os.path.basename(path)
pattern = r'\.(\d+)\.'
matches = list(re.finditer(pattern, name) or [])
if not matches:
return path
# Get last match.
match = matches[-1]
frame_token = token * len(match.group(1))
start, end = match.span()
apetail_name = '%s.%s.%s' % (name[:start], frame_token, name[end:])
return os.path.join(folder, apetail_name)[::-1]
# Success
eg1 = 'xx01_010_animation.0100.exr'
eg1 = sub_frame_number_for_frame_token(eg1) # result: xx01_010_animation.####.exr
# Failure
eg2 = 'xx01_010_animation.123.0100.exr'
eg2 = sub_frame_number_for_frame_token(eg2) # result: xx01_010_animation.123.####.exr
print(eg1)
print(eg2)

I believe the problem is that finditer returns only non-overlapping matches. Because both '.' characters are part of the regular expression, it doesn't consider the second dot as a possible start of another match. You can probably use the lookahead construct ?= to match the second dot without consuming it with "?=.".
Because of the way regular expressions work, I don't think there is an easy way to search right-to-left (though I suppose you could reverse the string and write the pattern backwards...).

If all you care about is the last \.(\d+)\., then anchor your pattern from the end of the string and do a simple re.search(_):
\.(\d+)\.(?:.*?)$
where (?:.*?) is non-capturing and non-greedy, so it will consume as few characters as possible between your real target and the end of the string, and those characters will not show up in matches.
(Caveat 1: I have not tested this. Caveat 2: That is one ugly regex, so add a comment explaining what it's doing.)
UPDATE: Actually I guess you could just do a ^.*(\.\d\.) and let the implicitly greedy .* match as much as possible (including matches that occur earlier in the string) while still matching your group. That makes for a simpler regex, but I think it makes your intentions less clear.

python regular expression to match strings

I want to parse a string, such as:
package: name='jp.tjkapp.droid1lwp' versionCode='2' versionName='1.1'
uses-permission:'android.permission.WRITE_APN_SETTINGS'
uses-permission:'android.permission.RECEIVE_BOOT_COMPLETED'
uses-permission:'android.permission.ACCESS_NETWORK_STATE'
I want to get:
string1: jp.tjkapp.droidllwp`
string2: 1.1
Because there are multiple uses-permission, I want to get permission as a list, contains:
WRITE_APN_SETTINGS, RECEIVE_BOOT_COMPLETED and ACCESS_NETWORK_STATE.
Could you help me write the python regular expression to get the strings I want?
Thanks.

Assuming the code block you provided is one long string, here stored in a variable called input_string:
name = re.search(r"(?<=name\=\')[\w\.]+?(?=\')", input_string).group(0)
versionName = re.search(r"(?<=versionName\=\')\d+?\.\d+?(?=\')", input_string).group(0)
permissions = re.findall(r'(?<=android\.permission\.)[A-Z_]+(?=\')', input_string)
Explanation:
name
(?<=name\=\'): check ahead of the main string in order to return only strings that are preceded by name='. The \ in front of = and ' serve to escape them so that the regex knows we're talking about the = string and not a regex command. name=' is not also returned when we get the result, we just know that the results we get are all preceded by it.
[\w\.]+?: This is the main string we're searching for. \w means any alphanumeric character and underscore. \. is an escaped period, so the regex knows we mean . and not the regex command represented by an unescaped period. Putting these in [] means we're okay with anything we've stuck in brackets, so we're saying that we'll accept any alphanumeric character, _, or .. + afterwords means at least one of the previous thing, meaning at least one (but possibly more) of [\w\.]. Finally, the ? means don't be greedy--we're telling the regex to get the smallest possible group that meets these specifications, since + could go on for an unlimited number of repeats of anything matched by [\w\.].
(?=\'): check behind the main string in order to return only strings that are followed by '. The \ is also an escape, since otherwise regex or Python's string execution might misinterpret '. This final ' is not returned with our results, we just know that in the original string, it followed any result we do end up getting.

You can do this without regex by reading the file content line by line.
>>> def split_string(s):
... if s.startswith('package'):
... return [i.split('=')[1] for i in s.split() if "=" in i]
... elif s.startswith('uses-permission'):
... return s.split('.')[-1]
...
>>> split_string("package: name='jp.tjkapp.droid1lwp' versionCode='2' versionName='1.1'")
["'jp.tjkapp.droid1lwp'", "'2'", "'1.1'"]
>>> split_string("uses-permission:'android.permission.WRITE_APN_SETTINGS'")
"WRITE_APN_SETTINGS'"
>>> split_string("uses-permission:'android.permission.RECEIVE_BOOT_COMPLETED'")
"RECEIVE_BOOT_COMPLETED'"
>>> split_string("uses-permission:'android.permission.ACCESS_NETWORK_STATE'")
"ACCESS_NETWORK_STATE'"
>>>

Here is one example code
#!/usr/bin/env python
inputFile = open("test.txt", "r").readlines()
for line in inputFile:
if line.startswith("package"):
words = line.split()
string1 = words[1].split("=")[1].replace("'","")
string2 = words[3].split("=")[1].replace("'","")
test.txt file contains input data you mentioned earlier..

python regex for repeating string

I am wanting to verify and then parse this string (in quotes):
string = "start: c12354, c3456, 34526; other stuff that I don't care about"
//Note that some codes begin with 'c'
I would like to verify that the string starts with 'start:' and ends with ';'
Afterward, I would like to have a regex parse out the strings. I tried the following python re code:
regx = r"start: (c?[0-9]+,?)+;"
reg = re.compile(regx)
matched = reg.search(string)
print ' matched.groups()', matched.groups()
I have tried different variations but I can either get the first or the last code but not a list of all three.
Or should I abandon using a regex?
EDIT: updated to reflect part of the problem space I neglected and fixed string difference.
Thanks for all the suggestions - in such a short time.

In Python, this isn’t possible with a single regular expression: each capture of a group overrides the last capture of that same group (in .NET, this would actually be possible since the engine distinguishes between captures and groups).
Your easiest solution is to first extract the part between start: and ; and then using a regular expression to return all matches, not just a single match, using re.findall('c?[0-9]+', text).

You could use the standard string tools, which are pretty much always more readable.
s = "start: c12354, c3456, 34526;"
s.startswith("start:") # returns a boolean if it starts with this string
s.endswith(";") # returns a boolean if it ends with this string
s[6:-1].split(', ') # will give you a list of tokens separated by the string ", "

This can be done (pretty elegantly) with a tool like Pyparsing:
from pyparsing import Group, Literal, Optional, Word
import string
code = Group(Optional(Literal("c"), default='') + Word(string.digits) + Optional(Literal(","), default=''))
parser = Literal("start:") + OneOrMore(code) + Literal(";")
# Read lines from file:
with open('lines.txt', 'r') as f:
for line in f:
try:
result = parser.parseString(line)
codes = [c[1] for c in result[1:-1]]
# Do something with teh codez...
except ParseException exc:
# Oh noes: string doesn't match!
continue
Cleaner than a regular expression, returns a list of codes (no need to string.split), and ignores any extra characters in the line, just like your example.

import re
sstr = re.compile(r'start:([^;]*);')
slst = re.compile(r'(?:c?)(\d+)')
mystr = "start: c12354, c3456, 34526; other stuff that I don't care about"
match = re.match(sstr, mystr)
if match:
res = re.findall(slst, match.group(0))
results in
['12354', '3456', '34526']

finding and returning a string with a specified prefix

I am close but I am not sure what to do with the restuling match object. If I do
p = re.search('[/#.* /]', str)
I'll get any words that start with # and end up with a space. This is what I want. However this returns a Match object that I dont' know what to do with. What's the most computationally efficient way of finding and returning a string which is prefixed with a #?
For example,
"Hi there #guy"
After doing the proper calculations, I would be returned
guy

The following regular expression do what you need:
import re
s = "Hi there #guy"
p = re.search(r'#(\w+)', s)
print p.group(1)
It will also work for the following string formats:
s = "Hi there #guy " # notice the trailing space
s = "Hi there #guy," # notice the trailing comma
s = "Hi there #guy and" # notice the next word
s = "Hi there #guy22" # notice the trailing numbers
s = "Hi there #22guy" # notice the leading numbers

That regex does not do what you think it does.
s = "Hi there #guy"
p = re.search(r'#([^ ]+)', s) # this is the regex you described
print p.group(1) # first thing matched inside of ( .. )
But as usually with regex, there are tons of examples that break this, for example if the text is s = "Hi there #guy, what's with the comma?" the result would be guy,.
So you really need to think about every possible thing you want and don't want to match. r'#([a-zA-Z]+)' might be a good starting point, it literally only matches letters (a .. z, no unicode etc).

p.group(0) should return guy. If you want to find out what function an object has, you can use the dir(p) method to find out. This will return a list of attributes and methods that are available for that object instance.

As it's evident from the answers so far regex is the most efficient solution for your problem. Answers differ slightly regarding what you allow to be followed by the #:
[^ ] anything but space
\w in python-2.x is equivalent to [A-Za-z0-9_], in py3k is locale dependent
If you have better idea what characters might be included in the user name you might adjust your regex to reflect that, e.g., only lower case ascii letters, would be:
[a-z]
NB: I skipped quantifiers for simplicity.

(?<=#)\w+
will match a word if it's preceded by a # (without adding it to the match, a so-called positive lookbehind). This will match "words" that are composed of letters, numbers, and/or underscore; if you don't want those, use (?<=#)[^\W\d_]+
In Python:
>>> strg = "Hi there #guy!"
>>> p = re.search(r'(?<=#)\w+', strg)
>>> p.group()
'guy'

You say: """If I do p = re.search('[/#.* /]', str) I'll get any words that start with # and end up with a space."" But this is incorrect -- that pattern is a character class which will match ONE character in the set #/.* and space. Note: there's a redundant second / in the pattern.
For example:
>>> re.findall('[/#.* /]', 'xxx#foo x/x.x*x xxxx')
['#', ' ', '/', '.', '*', ' ']
>>>
You say that you want "guy" returned from "Hi there #guy" but that conflicts with "and end up with a space".
Please edit your question to include what you really want/need to match.