introduce mutation variants to generic token analyser

Mutations are regular-expression-based replacements that are applied
after variants have been computed. They are meant to be used for
variations on character level.

Add spelling variations for German umlauts.

diff --git a/nominatim/tokenizer/token_analysis/generic.py b/nominatim/tokenizer/token_analysis/generic.py
index b9bd9bdf8d6043794a63bb104b44ac1040954841..1e7b75a1218ba8330a99d17d6681fa3c8c641b82 100644 (file)
--- a/nominatim/tokenizer/token_analysis/generic.py
+++ b/nominatim/tokenizer/token_analysis/generic.py
@@ -11,7 +11,9 @@ import itertools
 
 import datrie
 
 
 import datrie
 

+from nominatim.errors import UsageError

 from nominatim.tokenizer.token_analysis.config_variants import get_variant_config
 from nominatim.tokenizer.token_analysis.config_variants import get_variant_config

+from nominatim.tokenizer.token_analysis.generic_mutation import MutationVariantGenerator

 
 ### Configuration section
 
 
 ### Configuration section
 


@@ -23,6 +25,7 @@ def configure(rules, normalization_rules):
     config['replacements'], config['chars'] = get_variant_config(rules.get('variants'),
                                                                  normalization_rules)
     config['variant_only'] = rules.get('mode', '') == 'variant-only'
     config['replacements'], config['chars'] = get_variant_config(rules.get('variants'),
                                                                  normalization_rules)
     config['variant_only'] = rules.get('mode', '') == 'variant-only'

+    config['mutations'] = rules.get('mutations', [])

 
     return config
 
 
     return config
 


@@ -52,19 +55,45 @@ class GenericTokenAnalysis:
         else:
             self.replacements = None
 
         else:
             self.replacements = None
 

+        # set up mutation rules
+        self.mutations = []
+        for cfg in config['mutations']:
+            if 'pattern' not in cfg:
+                raise UsageError("Missing field 'pattern' in mutation configuration.")
+            if not isinstance(cfg['pattern'], str):
+                raise UsageError("Field 'pattern' in mutation configuration "
+                                 "must be a simple text field.")
+            if 'replacements' not in cfg:
+                raise UsageError("Missing field 'replacements' in mutation configuration.")
+            if not isinstance(cfg['replacements'], list):
+                raise UsageError("Field 'replacements' in mutation configuration "
+                                 "must be a list of texts.")
+
+            self.mutations.append(MutationVariantGenerator(cfg['pattern'],
+                                                           cfg['replacements']))
+

 
     def get_variants_ascii(self, norm_name):
         """ Compute the spelling variants for the given normalized name
             and transliterate the result.
         """
 
     def get_variants_ascii(self, norm_name):
         """ Compute the spelling variants for the given normalized name
             and transliterate the result.
         """

-        results = set()
-        for variant in self._generate_word_variants(norm_name):
-            if not self.variant_only or variant.strip() != norm_name:
-                trans_name = self.to_ascii.transliterate(variant).strip()
-                if trans_name:
-                    results.add(trans_name)
-
-        return list(results)
+        variants = self._generate_word_variants(norm_name)
+
+        for mutation in self.mutations:
+            variants = mutation.generate(variants)
+
+        return [name for name in self._transliterate_unique_list(norm_name, variants) if name]
+
+
+    def _transliterate_unique_list(self, norm_name, iterable):
+        seen = set()
+        if self.variant_only:
+            seen.add(norm_name)
+
+        for variant in map(str.strip, iterable):
+            if variant not in seen:
+                seen.add(variant)
+                yield self.to_ascii.transliterate(variant).strip()

 
 
     def _generate_word_variants(self, norm_name):
 
 
     def _generate_word_variants(self, norm_name):

diff --git a/nominatim/tokenizer/token_analysis/generic_mutation.py b/nominatim/tokenizer/token_analysis/generic_mutation.py
new file mode 100644 (file)
index 0000000..d23d5cd
--- /dev/null
+++ b/nominatim/tokenizer/token_analysis/generic_mutation.py
@@ -0,0 +1,56 @@
+# SPDX-License-Identifier: GPL-2.0-only
+#
+# This file is part of Nominatim. (https://nominatim.org)
+#
+# Copyright (C) 2022 by the Nominatim developer community.
+# For a full list of authors see the git log.
+"""
+Creator for mutation variants for the generic token analysis.
+"""
+import itertools
+import logging
+import re
+
+from nominatim.errors import UsageError
+
+LOG = logging.getLogger()
+
+def _zigzag(outer, inner):
+    return itertools.chain.from_iterable(itertools.zip_longest(outer, inner, fillvalue=''))
+
+
+class MutationVariantGenerator:
+    """ Generates name variants by applying a regular expression to the name
+        and replacing it with one or more variants. When the regular expression
+        matches more than once, each occurence is replaced with all replacement
+        patterns.
+    """
+
+    def __init__(self, pattern, replacements):
+        self.pattern = re.compile(pattern)
+        self.replacements = replacements
+
+        if self.pattern.groups > 0:
+            LOG.fatal("The mutation pattern %s contains a capturing group. "
+                      "This is not allowed.", pattern)
+            raise UsageError("Bad mutation pattern in configuration.")
+
+
+    def generate(self, names):
+        """ Generator function for the name variants. 'names' is an iterable
+            over a set of names for which the variants are to be generated.
+        """
+        for name in names:
+            parts = self.pattern.split(name)
+            if len(parts) == 1:
+                yield name
+            else:
+                for seps in self._fillers(len(parts)):
+                    yield ''.join(_zigzag(parts, seps))
+
+
+    def _fillers(self, num_parts):
+        """ Returns a generator for strings to join the given number of string
+            parts in all possible combinations.
+        """
+        return itertools.product(self.replacements, repeat=num_parts - 1)

diff --git a/settings/icu_tokenizer.yaml b/settings/icu_tokenizer.yaml
index a3c62e67b8d2a500baee37223f56b151bce14324..c6601faf0645186740c06b99eaec124744aba0e0 100644 (file)
--- a/settings/icu_tokenizer.yaml
+++ b/settings/icu_tokenizer.yaml
@@ -59,6 +59,13 @@ token-analysis:
       mode: variant-only
       variants:
           - !include icu-rules/variants-de.yaml
       mode: variant-only
       variants:
           - !include icu-rules/variants-de.yaml

+      mutations:
+          - pattern: ä
+            replacements: ["ä", "ae"]
+          - pattern: ö
+            replacements: ["ö", "oe"]
+          - pattern: ü
+            replacements: ["ü", "ue"]

     - id: el
       analyzer: generic
       mode: variant-only
     - id: el
       analyzer: generic
       mode: variant-only

diff --git a/test/bdd/db/import/naming.feature b/test/bdd/db/import/naming.feature
index bb29d2a314a46f7f917e298c4cf37fa311eb340c..b739cbae0482a65f0824f35b231012bdde387dde 100644 (file)
--- a/test/bdd/db/import/naming.feature
+++ b/test/bdd/db/import/naming.feature
@@ -58,3 +58,48 @@ Feature: Import and search of names
         | រាជធានីភ្នំពេញ |
         | 東京都 |
         | ပုဗ္ဗသီရိ |
         | រាជធានីភ្នំពេញ |
         | 東京都 |
         | ပုဗ္ဗသီရိ |

+
+
+    Scenario: German umlauts can be found when expanded
+        Given the places
+            | osm | class | type | name+name:de |
+            | N1  | place | city | Münster      |
+            | N2  | place | city | Köln         |
+            | N3  | place | city | Gräfenroda   |
+        When importing
+        When sending search query "münster"
+        Then results contain
+            | osm |
+            | N1  |
+        When sending search query "muenster"
+        Then results contain
+            | osm |
+            | N1  |
+        When sending search query "munster"
+        Then results contain
+            | osm |
+            | N1  |
+        When sending search query "Köln"
+        Then results contain
+            | osm |
+            | N2  |
+        When sending search query "Koeln"
+        Then results contain
+            | osm |
+            | N2  |
+        When sending search query "Koln"
+        Then results contain
+            | osm |
+            | N2  |
+        When sending search query "gräfenroda"
+        Then results contain
+            | osm |
+            | N3  |
+        When sending search query "graefenroda"
+        Then results contain
+            | osm |
+            | N3  |
+        When sending search query "grafenroda"
+        Then results contain
+            | osm |
+            | N3  |

diff --git a/test/python/tokenizer/token_analysis/test_generic_mutation.py b/test/python/tokenizer/token_analysis/test_generic_mutation.py
new file mode 100644 (file)
index 0000000..757f031
--- /dev/null
+++ b/test/python/tokenizer/token_analysis/test_generic_mutation.py
@@ -0,0 +1,89 @@
+# SPDX-License-Identifier: GPL-2.0-only
+#
+# This file is part of Nominatim. (https://nominatim.org)
+#
+# Copyright (C) 2022 by the Nominatim developer community.
+# For a full list of authors see the git log.
+"""
+Tests for generic token analysis, mutation part.
+"""
+import pytest
+
+from icu import Transliterator
+
+import nominatim.tokenizer.token_analysis.generic as module
+from nominatim.errors import UsageError
+
+DEFAULT_NORMALIZATION = """ '🜳' > ' ';
+                            [[:Nonspacing Mark:] [:Cf:]] >;
+                            :: lower ();
+                            [[:Punctuation:][:Space:]]+ > ' '
+                        """
+
+DEFAULT_TRANSLITERATION = """ ::  Latin ();
+                              '🜵' > ' ';
+                          """
+
+class TestMutationNoVariants:
+
+    def make_analyser(self, *mutations):
+        rules = { 'analyzer': 'generic',
+                  'mutations': [ {'pattern': m[0], 'replacements': m[1]}
+                                 for m in mutations]
+                }
+        config = module.configure(rules, DEFAULT_NORMALIZATION)
+        trans = Transliterator.createFromRules("test_trans", DEFAULT_TRANSLITERATION)
+
+        self.analysis = module.create(trans, config)
+
+
+    def variants(self, name):
+        norm = Transliterator.createFromRules("test_norm", DEFAULT_NORMALIZATION)
+        return set(self.analysis.get_variants_ascii(norm.transliterate(name).strip()))
+
+
+    @pytest.mark.parametrize('pattern', ('(capture)', ['a list']))
+    def test_bad_pattern(self, pattern):
+        with pytest.raises(UsageError):
+            self.make_analyser((pattern, ['b']))
+
+
+    @pytest.mark.parametrize('replacements', (None, 'a string'))
+    def test_bad_replacement(self, replacements):
+        with pytest.raises(UsageError):
+            self.make_analyser(('a', replacements))
+
+
+    def test_simple_replacement(self):
+        self.make_analyser(('a', ['b']))
+
+        assert self.variants('none') == {'none'}
+        assert self.variants('abba') == {'bbbb'}
+        assert self.variants('2 aar') == {'2 bbr'}
+
+
+    def test_multichar_replacement(self):
+        self.make_analyser(('1 1', ['1 1 1']))
+
+        assert self.variants('1 1456') == {'1 1 1456'}
+        assert self.variants('1 1 1') == {'1 1 1 1'}
+
+
+    def test_removement_replacement(self):
+        self.make_analyser((' ', [' ', '']))
+
+        assert self.variants('A 345') == {'a 345', 'a345'}
+        assert self.variants('a g b') == {'a g b', 'ag b', 'a gb', 'agb'}
+
+
+    def test_regex_pattern(self):
+        self.make_analyser(('[^a-z]+', ['XXX', ' ']))
+
+        assert self.variants('a-34n12') == {'aXXXnXXX', 'aXXXn', 'a nXXX', 'a n'}
+
+
+    def test_multiple_mutations(self):
+        self.make_analyser(('ä', ['ä', 'ae']), ('ö', ['ö', 'oe']))
+
+        assert self.variants('Längenöhr') == {'längenöhr', 'laengenöhr',
+                                              'längenoehr', 'laengenoehr'}