Extracting info from large structured text files

Question

I need to read some large files (from 50k to 100k lines), structured in groups separated by empty lines. Each group start at the same pattern "No.999999999 dd/mm/yyyy ZZZ". Here´s some sample data.

No.813829461 16/09/1987 270
Tit.SUZANO PAPEL E CELULOSE S.A. (BR/BA)
C.N.P.J./C.I.C./N INPI : 16404287000155
Procurador: MARCELLO DO NASCIMENTO

No.815326777 28/12/1989 351
Tit.SIGLA SISTEMA GLOBO DE GRAVACOES AUDIO VISUAIS LTDA (BR/RJ)
C.N.P.J./C.I.C./NºINPI : 34162651000108
Apres.: Nominativa ; Nat.: De Produto
Marca: TRIO TROPICAL
Clas.Prod/Serv: 09.40
*DEFERIDO CONFORME RESOLUÇÃO 123 DE 06/01/2006, PUBLICADA NA RPI 1829, DE 24/01/2006.
Procurador: WALDEMAR RODRIGUES PEDRA

No.900148764 11/01/2007 LD3
Tit.TIARA BOLSAS E CALÇADOS LTDA
Procurador: Marcia Ferreira Gomes
*Escritório: Marcas Marcantes e Patentes Ltda
*Exigência Formal não respondida Satisfatoriamente, Pedido de Registro de Marca considerado inexistente, de acordo com Art. 157 da LPI
*Protocolo da Petição de cumprimento de Exigência Formal: 810080140197

I wrote some code that´s parsing it accordingly. There´s anything that I can improve, to improve readability or performance? Here´s what I come so far:

import re, pprint

class Despacho(object):
    """
    Class to parse each line, applying the regexp and storing the results
    for future use
    """
    regexp = {
        re.compile(r'No.([\d]{9})  ([\d]{2}/[\d]{2}/[\d]{4})  (.*)'): lambda self: self._processo,
        re.compile(r'Tit.(.*)'): lambda self: self._titular,
        re.compile(r'Procurador: (.*)'): lambda self: self._procurador,
        re.compile(r'C.N.P.J./C.I.C./N INPI :(.*)'): lambda self: self._documento,
        re.compile(r'Apres.: (.*) ; Nat.: (.*)'): lambda self: self._apresentacao,
        re.compile(r'Marca: (.*)'): lambda self: self._marca,
        re.compile(r'Clas.Prod/Serv: (.*)'): lambda self: self._classe,
        re.compile(r'\*(.*)'): lambda self: self._complemento,
    }

    def __init__(self):
        """
        'complemento' is the only field that can be multiple in a single registry
        """
        self.complemento = []

    def _processo(self, matches):
        self.processo, self.data, self.despacho = matches.groups()

    def _titular(self, matches):
        self.titular = matches.group(1)

    def _procurador(self, matches):
        self.procurador = matches.group(1)

    def _documento(self, matches):
        self.documento = matches.group(1)

    def _apresentacao(self, matches):
        self.apresentacao, self.natureza = matches.groups()

    def _marca(self, matches):
        self.marca = matches.group(1)

    def _classe(self, matches):
        self.classe = matches.group(1)

    def _complemento(self, matches):
        self.complemento.append(matches.group(1))

    def read(self, line):
        for pattern in Despacho.regexp:
            m = pattern.match(line)
            if m:
                Despacho.regexp[pattern](self)(m)


def process(rpi):
    """
    read data and process each group
    """
    rpi = (line for line in rpi)
    group = False

    for line in rpi:
        if line.startswith('No.'):
            group = True
            d = Despacho()        

        if not line.strip() and group: # empty line - end of block
            yield d
            group = False

        d.read(line)


arquivo = open('rm1972.txt') # file to process
for desp in process(arquivo):
    pprint.pprint(desp.__dict__)
    print('--------------')

Thank you.

By the way, sorry for my poor english. :)

Answer 1

It would be easier to help if you had a specific concern. Performance will depend greatly on the efficiency of the particular regex engine you are using. 100K lines in a single file doesn't sound that big, but again it all depends on your environment.

I use Expresso in my .NET development to test expressions for accuracy and performance. A Google search turned up Kodos, a GUI Python regex authoring tool.

Answer 2

I'm not going to go over everything but for starters: instead of making functions that each of the lambdas call, why not perform the work directly in the lambda?

For example:

    regexp = {
            re.compile(r'No.([\d]{9})  ([\d]{2}/[\d]{2}/[\d]{4})  (.*)'): lambda self, matches: self.processo, self.data, self.despacho = matches.groups()
    }

...

Despacho.regexp[pattern](self, m)

That saves you a lot of function definitions.

Answer 3

It looks good overall, but why do you have the line:

rpi = (line for line in rpi)

You can already iterate over the file object without this intermediate step.

Answer 4

That is pretty good. Below some suggestions, let me know if you like'em:

import re
import pprint
import sys

class Despacho(object):
    """
    Class to parse each line, applying the regexp and storing the results
    for future use
    """
    #used a dict with the keys instead of functions.
    regexp = {
        ('processo', 
         'data', 
         'despacho'): re.compile(r'No.([\d]{9})  ([\d]{2}/[\d]{2}/[\d]{4})  (.*)'),
        ('titular',): re.compile(r'Tit.(.*)'),
        ('procurador',): re.compile(r'Procurador: (.*)'),
        ('documento',): re.compile(r'C.N.P.J./C.I.C./N INPI :(.*)'),
        ('apresentacao',
         'natureza'): re.compile(r'Apres.: (.*) ; Nat.: (.*)'),
        ('marca',): re.compile(r'Marca: (.*)'),
        ('classe',): re.compile(r'Clas.Prod/Serv: (.*)'),
        ('complemento',): re.compile(r'\*(.*)'),
    }

    def __init__(self):
        """
        'complemento' is the only field that can be multiple in a single registry
        """
        self.complemento = []


    def read(self, line):
        for attrs, pattern in Despacho.regexp.iteritems():
            m = pattern.match(line)
            if m:
                for groupn, attr in enumerate(attrs):
                    # special case complemento:
                    if attr == 'complemento':
                        self.complemento.append(m.group(groupn + 1))
                    else:
                        # set the attribute on the object
                        setattr(self, attr, m.group(groupn + 1))

    def __repr__(self):
        # defines object printed representation
        d = {}
        for attrs in self.regexp:
            for attr in attrs:
                d[attr] = getattr(self, attr, None)
        return pprint.pformat(d)

def process(rpi):
    """
    read data and process each group
    """
    #Useless line, since you're doing a for anyway
    #rpi = (line for line in rpi)
    group = False

    for line in rpi:
        if line.startswith('No.'):
            group = True
            d = Despacho()        

        if not line.strip() and group: # empty line - end of block
            yield d
            group = False

        d.read(line)

def main():
    arquivo = open('rm1972.txt') # file to process
    for desp in process(arquivo):
        print desp # can print directly here.
        print('-' * 20)
    return 0

if __name__ == '__main__':
    main()

Answer 5

I wouldn't use regex here. If you know that your lines will be starting with fixed strings, why not check those strings and write a logic around it?

for line in open(file):
    if line[0:3]=='No.':
        currIndex='No'
        map['No']=line[4:]
   ....
   ...
   else if line.strip()=='':
       //store the record in the map and clear the map
   else:
      //append line to the last index in map.. this is when the record overflows to the next line.
      Map[currIndex]=Map[currIndex]+"\n"+line

Consider the above code as just the pseudocode.

Answer 6

Another version with only one combined regular expression:

#!/usr/bin/python

import re
import pprint
import sys

class Despacho(object):
    """
    Class to parse each line, applying the regexp and storing the results
    for future use
    """
    #used a dict with the keys instead of functions.
    regexp = re.compile(
        r'No.(?P<processo>[\d]{9})  (?P<data>[\d]{2}/[\d]{2}/[\d]{4})  (?P<despacho>.*)'
        r'|Tit.(?P<titular>.*)'
        r'|Procurador: (?P<procurador>.*)'
        r'|C.N.P.J./C.I.C./N INPI :(?P<documento>.*)'
        r'|Apres.: (?P<apresentacao>.*) ; Nat.: (?P<natureza>.*)'
        r'|Marca: (?P<marca>.*)'
        r'|Clas.Prod/Serv: (?P<classe>.*)'
        r'|\*(?P<complemento>.*)')

    simplefields = ('processo', 'data', 'despacho', 'titular', 'procurador',
                    'documento', 'apresentacao', 'natureza', 'marca', 'classe')

    def __init__(self):
        """
        'complemento' is the only field that can be multiple in a single
        registry
        """
        self.__dict__ = dict.fromkeys(self.simplefields)
        self.complemento = []

    def parse(self, line):
        m = self.regexp.match(line)
        if m:
            gd = dict((k, v) for k, v in m.groupdict().items() if v)
            if 'complemento' in gd:
                self.complemento.append(gd['complemento'])
            else:
                self.__dict__.update(gd)

    def __repr__(self):
        # defines object printed representation
        return pprint.pformat(self.__dict__)

def process(rpi):
    """
    read data and process each group
    """
    d = None

    for line in rpi:
        if line.startswith('No.'):
            if d:
                yield d
            d = Despacho()
        d.parse(line)
    yield d

def main():
    arquivo = file('rm1972.txt') # file to process
    for desp in process(arquivo):
        print desp # can print directly here.
        print '-' * 20

if __name__ == '__main__':
    main()