]> git.openstreetmap.org Git - nominatim.git/blobdiff - nominatim/api/search/icu_tokenizer.py
Merge pull request #3385 from mtmail/docs-reverse-only
[nominatim.git] / nominatim / api / search / icu_tokenizer.py
index 17e679057ecb5a4eb63daeb08f38e361f19a3ada..eb90c122eb43277a18b2d71bf48eb9ab99a375ac 100644 (file)
@@ -8,7 +8,6 @@
 Implementation of query analysis for the ICU tokenizer.
 """
 from typing import Tuple, Dict, List, Optional, NamedTuple, Iterator, Any, cast
-from copy import copy
 from collections import defaultdict
 import dataclasses
 import difflib
@@ -22,6 +21,7 @@ from nominatim.api.connection import SearchConnection
 from nominatim.api.logging import log
 from nominatim.api.search import query as qmod
 from nominatim.api.search.query_analyzer_factory import AbstractQueryAnalyzer
+from nominatim.db.sqlalchemy_types import Json
 
 
 DB_TO_TOKEN_TYPE = {
@@ -83,7 +83,7 @@ class ICUToken(qmod.Token):
         seq = difflib.SequenceMatcher(a=self.lookup_word, b=norm)
         distance = 0
         for tag, afrom, ato, bfrom, bto in seq.get_opcodes():
-            if tag == 'delete' and (afrom == 0 or ato == len(self.lookup_word)):
+            if tag in ('delete', 'insert') and (afrom == 0 or ato == len(self.lookup_word)):
                 distance += 1
             elif tag == 'replace':
                 distance += max((ato-afrom), (bto-bfrom))
@@ -97,14 +97,21 @@ class ICUToken(qmod.Token):
         """ Create a ICUToken from the row of the word table.
         """
         count = 1 if row.info is None else row.info.get('count', 1)
+        addr_count = 1 if row.info is None else row.info.get('addr_count', 1)
 
         penalty = 0.0
         if row.type == 'w':
             penalty = 0.3
+        elif row.type == 'W':
+            if len(row.word_token) == 1 and row.word_token == row.word:
+                penalty = 0.2 if row.word.isdigit() else 0.3
         elif row.type == 'H':
             penalty = sum(0.1 for c in row.word_token if c != ' ' and not c.isdigit())
             if all(not c.isdigit() for c in row.word_token):
                 penalty += 0.2 * (len(row.word_token) - 1)
+        elif row.type == 'C':
+            if len(row.word_token) == 1:
+                penalty = 0.3
 
         if row.info is None:
             lookup_word = row.word
@@ -115,9 +122,10 @@ class ICUToken(qmod.Token):
         else:
             lookup_word = row.word_token
 
-        return ICUToken(penalty=penalty, token=row.word_id, count=count,
+        return ICUToken(penalty=penalty, token=row.word_id, count=max(1, count),
                         lookup_word=lookup_word, is_indexed=True,
-                        word_token=row.word_token, info=row.info)
+                        word_token=row.word_token, info=row.info,
+                        addr_count=max(1, addr_count))
 
 
 
@@ -133,10 +141,19 @@ class ICUQueryAnalyzer(AbstractQueryAnalyzer):
     async def setup(self) -> None:
         """ Set up static data structures needed for the analysis.
         """
-        rules = await self.conn.get_property('tokenizer_import_normalisation')
-        self.normalizer = Transliterator.createFromRules("normalization", rules)
-        rules = await self.conn.get_property('tokenizer_import_transliteration')
-        self.transliterator = Transliterator.createFromRules("transliteration", rules)
+        async def _make_normalizer() -> Any:
+            rules = await self.conn.get_property('tokenizer_import_normalisation')
+            return Transliterator.createFromRules("normalization", rules)
+
+        self.normalizer = await self.conn.get_cached_value('ICUTOK', 'normalizer',
+                                                           _make_normalizer)
+
+        async def _make_transliterator() -> Any:
+            rules = await self.conn.get_property('tokenizer_import_transliteration')
+            return Transliterator.createFromRules("transliteration", rules)
+
+        self.transliterator = await self.conn.get_cached_value('ICUTOK', 'transliterator',
+                                                               _make_transliterator)
 
         if 'word' not in self.conn.t.meta.tables:
             sa.Table('word', self.conn.t.meta,
@@ -144,7 +161,7 @@ class ICUQueryAnalyzer(AbstractQueryAnalyzer):
                      sa.Column('word_token', sa.Text, nullable=False),
                      sa.Column('type', sa.Text, nullable=False),
                      sa.Column('word', sa.Text),
-                     sa.Column('info', self.conn.t.types.Json))
+                     sa.Column('info', Json))
 
 
     async def analyze_query(self, phrases: List[qmod.Phrase]) -> qmod.QueryStruct:
@@ -153,7 +170,7 @@ class ICUQueryAnalyzer(AbstractQueryAnalyzer):
         """
         log().section('Analyze query (using ICU tokenizer)')
         normalized = list(filter(lambda p: p.text,
-                                 (qmod.Phrase(p.ptype, self.normalizer.transliterate(p.text))
+                                 (qmod.Phrase(p.ptype, self.normalize_text(p.text))
                                   for p in phrases)))
         query = qmod.QueryStruct(normalized)
         log().var_dump('Normalized query', query.source)
@@ -169,13 +186,12 @@ class ICUQueryAnalyzer(AbstractQueryAnalyzer):
                 if row.type == 'S':
                     if row.info['op'] in ('in', 'near'):
                         if trange.start == 0:
-                            query.add_token(trange, qmod.TokenType.CATEGORY, token)
+                            query.add_token(trange, qmod.TokenType.NEAR_ITEM, token)
                     else:
-                        query.add_token(trange, qmod.TokenType.QUALIFIER, token)
-                        if trange.start == 0 or trange.end == query.num_token_slots():
-                            token = copy(token)
-                            token.penalty += 0.1 * (query.num_token_slots())
-                            query.add_token(trange, qmod.TokenType.CATEGORY, token)
+                        if trange.start == 0 and trange.end == query.num_token_slots():
+                            query.add_token(trange, qmod.TokenType.NEAR_ITEM, token)
+                        else:
+                            query.add_token(trange, qmod.TokenType.QUALIFIER, token)
                 else:
                     query.add_token(trange, DB_TO_TOKEN_TYPE[row.type], token)
 
@@ -187,6 +203,14 @@ class ICUQueryAnalyzer(AbstractQueryAnalyzer):
         return query
 
 
+    def normalize_text(self, text: str) -> str:
+        """ Bring the given text into a normalized form. That is the
+            standardized form search will work with. All information removed
+            at this stage is inevitably lost.
+        """
+        return cast(str, self.normalizer.transliterate(text))
+
+
     def split_query(self, query: qmod.QueryStruct) -> Tuple[QueryParts, WordDict]:
         """ Transliterate the phrases and split them into tokens.
 
@@ -235,7 +259,7 @@ class ICUQueryAnalyzer(AbstractQueryAnalyzer):
             if len(part.token) <= 4 and part[0].isdigit()\
                and not node.has_tokens(i+1, qmod.TokenType.HOUSENUMBER):
                 query.add_token(qmod.TokenRange(i, i+1), qmod.TokenType.HOUSENUMBER,
-                                ICUToken(0.5, 0, 1, part.token, True, part.token, None))
+                                ICUToken(0.5, 0, 1, 1, part.token, True, part.token, None))
 
 
     def rerank_tokens(self, query: qmod.QueryStruct, parts: QueryParts) -> None:
@@ -248,12 +272,11 @@ class ICUQueryAnalyzer(AbstractQueryAnalyzer):
                        and (repl.ttype != qmod.TokenType.HOUSENUMBER
                             or len(tlist.tokens[0].lookup_word) > 4):
                         repl.add_penalty(0.39)
-            elif tlist.ttype == qmod.TokenType.HOUSENUMBER:
+            elif tlist.ttype == qmod.TokenType.HOUSENUMBER \
+                 and len(tlist.tokens[0].lookup_word) <= 3:
                 if any(c.isdigit() for c in tlist.tokens[0].lookup_word):
                     for repl in node.starting:
-                        if repl.end == tlist.end and repl.ttype != qmod.TokenType.HOUSENUMBER \
-                           and (repl.ttype != qmod.TokenType.HOUSENUMBER
-                                or len(tlist.tokens[0].lookup_word) <= 3):
+                        if repl.end == tlist.end and repl.ttype != qmod.TokenType.HOUSENUMBER:
                             repl.add_penalty(0.5 - tlist.tokens[0].penalty)
             elif tlist.ttype not in (qmod.TokenType.COUNTRY, qmod.TokenType.PARTIAL):
                 norm = parts[i].normalized