nominatim/tokenizer/icu_name_processor.py

   1 """
   2 Processor for names that are imported into the database based on the
   3 ICU library.
   4 """
   5 from collections import defaultdict
   6 import itertools
   7
   8 from icu import Transliterator
   9 import datrie
  10
  11 from nominatim.db.properties import set_property, get_property
  12 from nominatim.tokenizer import icu_variants as variants
  13
  14 DBCFG_IMPORT_NORM_RULES = "tokenizer_import_normalisation"
  15 DBCFG_IMPORT_TRANS_RULES = "tokenizer_import_transliteration"
  16 DBCFG_IMPORT_REPLACEMENTS = "tokenizer_import_replacements"
  17 DBCFG_SEARCH_STD_RULES = "tokenizer_search_standardization"
  18
  19
  20 class ICUNameProcessorRules:
  21     """ Data object that saves the rules needed for the name processor.
  22
  23         The rules can either be initialised through an ICURuleLoader or
  24         be loaded from a database when a connection is given.
  25     """
  26     def __init__(self, loader=None, conn=None):
  27         if loader is not None:
  28             self.norm_rules = loader.get_normalization_rules()
  29             self.trans_rules = loader.get_transliteration_rules()
  30             self.replacements = loader.get_replacement_pairs()
  31             self.search_rules = loader.get_search_rules()
  32         elif conn is not None:
  33             self.norm_rules = get_property(conn, DBCFG_IMPORT_NORM_RULES)
  34             self.trans_rules = get_property(conn, DBCFG_IMPORT_TRANS_RULES)
  35             self.replacements = \
  36                 variants.unpickle_variant_set(get_property(conn, DBCFG_IMPORT_REPLACEMENTS))
  37             self.search_rules = get_property(conn, DBCFG_SEARCH_STD_RULES)
  38         else:
  39             assert False, "Parameter loader or conn required."
  40
  41
  42     def save_rules(self, conn):
  43         """ Save the rules in the property table of the given database.
  44             the rules can be loaded again by handing in a connection into
  45             the constructor of the class.
  46         """
  47         set_property(conn, DBCFG_IMPORT_NORM_RULES, self.norm_rules)
  48         set_property(conn, DBCFG_IMPORT_TRANS_RULES, self.trans_rules)
  49         set_property(conn, DBCFG_IMPORT_REPLACEMENTS,
  50                      variants.pickle_variant_set(self.replacements))
  51         set_property(conn, DBCFG_SEARCH_STD_RULES, self.search_rules)
  52
  53
  54 class ICUNameProcessor:
  55     """ Collects the different transformation rules for normalisation of names
  56         and provides the functions to aply the transformations.
  57     """
  58
  59     def __init__(self, rules):
  60         self.normalizer = Transliterator.createFromRules("icu_normalization",
  61                                                          rules.norm_rules)
  62         self.to_ascii = Transliterator.createFromRules("icu_to_ascii",
  63                                                        rules.trans_rules +
  64                                                        ";[:Space:]+ > ' '")
  65         self.search = Transliterator.createFromRules("icu_search",
  66                                                      rules.search_rules)
  67
  68         # Intermediate reorder by source. Also compute required character set.
  69         immediate = defaultdict(list)
  70         chars = set()
  71         for variant in rules.replacements:
  72             if variant.source[-1] == ' ' and variant.replacement[-1] == ' ':
  73                 replstr = variant.replacement[:-1]
  74             else:
  75                 replstr = variant.replacement
  76             immediate[variant.source].append(replstr)
  77             chars.update(variant.source)
  78         # Then copy to datrie
  79         self.replacements = datrie.Trie(''.join(chars))
  80         for src, repllist in immediate.items():
  81             self.replacements[src] = repllist
  82
  83
  84     def get_normalized(self, name):
  85         """ Normalize the given name, i.e. remove all elements not relevant
  86             for search.
  87         """
  88         return self.normalizer.transliterate(name).strip()
  89
  90     def get_variants_ascii(self, norm_name):
  91         """ Compute the spelling variants for the given normalized name
  92             and transliterate the result.
  93         """
  94         baseform = '^ ' + norm_name + ' ^'
  95         partials = ['']
  96
  97         startpos = 0
  98         pos = 0
  99         force_space = False
 100         while pos < len(baseform):
 101             full, repl = self.replacements.longest_prefix_item(baseform[pos:],
 102                                                                (None, None))
 103             if full is not None:
 104                 done = baseform[startpos:pos]
 105                 partials = [v + done + r
 106                             for v, r in itertools.product(partials, repl)
 107                             if not force_space or r.startswith(' ')]
 108                 startpos = pos + len(full)
 109                 if full[-1] == ' ':
 110                     startpos -= 1
 111                     force_space = True
 112                 pos = startpos
 113             else:
 114                 pos += 1
 115                 force_space = False
 116
 117         results = set()
 118
 119         if startpos == 0:
 120             trans_name = self.to_ascii.transliterate(norm_name).strip()
 121             if trans_name:
 122                 results.add(trans_name)
 123         else:
 124             for variant in partials:
 125                 name = variant + baseform[startpos:]
 126                 trans_name = self.to_ascii.transliterate(name[1:-1]).strip()
 127                 if trans_name:
 128                     results.add(trans_name)
 129
 130         return list(results)
 131
 132
 133     def get_search_normalized(self, name):
 134         """ Return the normalized version of the name (including transliteration)
 135             to be applied at search time.
 136         """
 137         return self.search.transliterate(' ' + name + ' ').strip()