pyrefo 0.3

a fast regex for object

pyrefo: a fast regex for object

This project is based on refo and the paper Regular Expression Matching: the Virtual Machine Approach, it use cffi to extend python with c to speed accelerate processing performance.

This project has done the following work:

1. full compatiable with refo api, support all patterns and match, search, finditer methods;
2. fix c source bug included in the paper;
3. use cffi to extend python with c;
4. add new feature which supports partial match;
5. add new Phrasepattern which can realize 'ab'match ['a', 'b', 'c']list;

How to use it

"ab" is Literal("a")+Literal("b")

"a*" is Star(Literal("a"))

"aab?" is Literal("a")+Literal("a")+Question(Literal("b"), greedy=False)

a{3,4} is Repetition(Literal("a"), 3, 4, greedy=False)

"(ab)+|(bb*)?" is

a = Literal("a")
b = Literal("b")
regex = Plus(a + b) | Star(b + b, greedy=False)

You can also assign a group to any sub-match and later on retrieve the matched content, for instance:

regex = Group(Plus(a + b), "foobar") | Star(b + b, greedy=False)
m = match(regex, "abab")
print(m.span("foobar"))

pyrefo offers match, search, findall, finditer search functions:

• match: match pattern from first position
• search: search pattern from first position till find one
• findall: find all matched result
• finditer: return an iterator for all matched result

pyrefo offers the following predicates:

• Any
• Literal
• Star
• Plus
• Question
• Group
• Repetition
• Phrase

Performance test

prerequisites

import jieba
text = '为什么在本店买东西？因为物流迅速＋品质保证。为什么我购买的每件商品评价都一样呢？因为我买的东西太多了，积累了很多未评价的订单，所以我统一用这段话作为评价内容。如果我用了这段话作为评价，那就说明这款产品非常赞，非常好！'
tokens = list(jieba.cut(text))

CPython

• pyrefo

from pyrefo import search, Group, Star, Any, Literal
%timeit search(Group(Literal('物流') + Star(Any()) + Literal('迅速'), 'a'), tokens)

95.9 µs ± 472 ns per loop (mean ± std. dev. of 7 runs, 10000 loops each)

• refo

import refo
%timeit refo.search(refo.Group(refo.Literal('物流') + refo.Star(refo.Any()) + refo.Literal('迅速'), 'a'), tokens)

1.03 ms ± 7.27 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)

• re

import re
%timeit re.search('(物流.*速度)', text)

989 ns ± 4.69 ns per loop (mean ± std. dev. of 7 runs, 1000000 loops each)

PyPy

• pyrefo

from pyrefo import search, Group, Star, Any, Literal
%timeit search(Group(Literal('物流') + Star(Any()) + Literal('迅速'), 'a'), tokens)

53.4 µs ± 28 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)

• refo

import refo
%timeit refo.search(refo.Group(refo.Literal('物流') + refo.Star(refo.Any()) + refo.Literal('迅速'), 'a'), tokens)

78 µs ± 35.8 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)

• re

import re
%timeit re.search('(物流.*速度)', text)

347 ns ± 3.26 ns per loop (mean ± std. dev. of 7 runs, 1000000 loops each)