seq-re

2-dimensional Sequence Regular Expression (SEQ RE)

This module provides regular expression matching operations over a sequence of tuples (or a sequence of sequence) data structure. It looks like the following::

seq_m_n = [[str_11, str_12, ... str_1n],
           [str_21, str_22, ... str_2n],
            ...,
           [str_m1, str_m2, ... str_mn]]

The sequence is a homogeneous 2D array, that is a matrix with m rows and n columns. In practice, m maybe vary from sequence to sequence, while n is usually a fixed-size.

A element in the tuple of the sequence can be considered as either a string, a word, a phrase, a char, a flag, a token or a tag, and maybe a set of tags or values (multi-values) in the future.

To match a pattern over a sequence of tuples, the SEQ RE patterns is written like one of the examples::

([;;PERSON]+) [was|has been] [an]? .{0,3} ([^painter|drawing artist|画家])

(?P<name@0,1,2>[;;PERSON]) [;VERB be;] [born] [on] (?P<birthday@0:3>([;;NUMBER|MONTH]|[-]){2,3})

• The syntax of SEQ RE pattern

A SEQ RE pattern is very similar to the ordinary regular express (RE) used in Python, in which the delimiters [...] is to indicate a tuple -- the second dimension of the sequence.

1.1 Inside [...] ++++++++++++++++++++

• [ and ]

  is the beginning and end delimiter of the tuple, e.g. [...].

• ;

  separates each element which the tuple contains, and the continuous ; at the tail can be omitted, e.g. [A|B;X;;], [A|B;X].

• |

  indicates the different values of one element, e.g. A|B. These values form a set, and any string in the set will be matched, e.g. A|B will match A or B.

• ^

  be the first character of an element, all the string that are not in the value set of this element will be matched. And ^ has no special meaning if it’s not the first character of the element. If ^ comes the first character of an element but it is a part of a literal string, \^ should be used to escape it.

• The priority of above-mentioned operations:

  [ ] < ; < ^ (not literal) < | < ^ (literal) .

• \

  is an escaping symbol before aforementioned special characters. Characters other than ], : or \ lose their special meaning inside [...]. To express ], : or | in literal, \ should be added before ], : or |. Meanwhile, to represent a literal backslash \ before ], ; or |, \\ should be used in the plain text that is to say '\\\\' must be used in the Python code.

1.2 Outside [...] +++++++++++++++++++++

• The special meanings of special characters in the ordinary RE are available here, but with the limitations discussed below.

  • Not support [ and ] as special characters to indicate a set of characters.

  • Not support the following escaped special characters: \number, \A, \b, \B, \d, \D, \s, \S, \w, \W, \Z, \a, \b, \f, \n, \r, \t, \v, \x.

  • Not support ranges of characters, such as [0-9A-Za-z], [\u4E00-\u9FBB\u3007] (Unihan and Chinese character 〇) used in ordinary RE.

  • The whitespace and non-special characters are ignored.

• . is an abbreviation of an arbitrary tuple [] or [;].

• The named groups in the pattern are very useful. As an extension, a format string starting with @ can be followed after the group name, to describe which element of the tuples belonging this group will be output as the result. For example: (?P<name@d1,d2:d3>...), in which d1, d2 and d3 are all 0-based position index number of elements in the tuple.

  • @0,2:4 means in the matched result only the 0th and from 2nd to 3rd elements of tuples will be output.

  • @@ means the pattern of the group itself will be output other than the matched result. one can choose whether to include the group name and parentheses or not.

  • @ means all elements of tuples in the matched result will be output.

1.3 Boolean logic in the [...] ++++++++++++++++++++++++++++++++++

Given a sequence of 3-tuple [[s1, s2, s3], ... ],

• AND

  [X;;Y] will match s1 == X && s3 == Y. Its behavior looks like the ordinary RE pattern (?:X.Y).

• OR

  [X;;]|[;;Y] will match s1 == X || s3 == Y. Its behavior looks like the ordinary RE pattern (?:X..)|(?:..Y)

• NOT

  If [;^P;] will match s2 != P. Its behavior looks like the ordinary RE pattern (?:.[^P].).

  We can also use a negative lookahead assertion of the ordinary RE, to give a negative covering its following. e.g. (?![;P;][Q])[;;][;;] <==> [;^P;][^Q;;], which behavior looks like the ordinary RE pattern (?!(?:.P.)(?:Q..))....

• Notes

Not support comparing the number of figures.

Multi-values of one element is not supported now, but this feature may be improved in the future.

Although SEQ RE has sufficient ability to express a pattern over sequences of tuples, it is still not a cascaded regular expressions (see also: Stanford TokensRegex <https://nlp.stanford.edu/software/tokensregex.html>_).

• Examples

The usage of seq_re module::

from __future__ import print_function
import seq_re

n = 3
pattern = ('(?P<name@0>[;;PERSON]+) [is|was|has been] [a|an]? '
           '(?P<attrib@0,1>.{0,3}) ([artist])')
seq = [['Vincent van Gogh', 'NNP', 'PERSON'],
       ['was', 'VBD', 'O'],
       ['a', 'DT', 'O'],
       ['Dutch', 'JJ', 'O'],
       ['Post-Impressionist', 'NN', 'O'],
       ['painter', 'NN', 'OCCUPATION'],
       ['who', 'WP', 'O'],
       ['is', 'VBZ', 'O'],
       ['among', 'IN', 'O'],
       ['the', 'DT', 'O'],
       ['most', 'RBS', 'O'],
       ['famous', 'JJ', 'O'],
       ['and', 'CC', 'O'],
       ['influential', 'JJ', 'O'],
       ['figures', 'NNS', 'O'],
       ['in', 'IN', 'O'],
       ['the', 'DT', 'O'],
       ['history', 'NN', 'O'],
       ['of', 'IN', 'O'],
       ['Western art', 'NNP', 'DOMAIN'],
       ['.', '.', 'O']]
placeholder_dict = {'artist': ['painter', 'drawing artist']}

sr = seq_re.SeqRegex(n).compile(pattern, **placeholder_dict)
match = sr.search(seq)
if match:
    for g in match.group_list:
        print(' '.join(['`'.join(tup) for tup in g[1]]))
    for name in sorted(match.named_group_dict,
                       key=lambda gn: match.named_group_dict[gn][0]):
        print(name, match.format_group_to_str(name, True))