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