]> git.openstreetmap.org Git - nominatim.git/blobdiff - nominatim/api/search/db_search_builder.py
Merge pull request #3375 from matkoniecz/patch-1
[nominatim.git] / nominatim / api / search / db_search_builder.py
index 62b4e1b51986b3a29becefee1078e969b5ad168f..e27a24d61eb54f0d7bb1bc04abedc90703dacc8a 100644 (file)
@@ -5,7 +5,7 @@
 # Copyright (C) 2023 by the Nominatim developer community.
 # For a full list of authors see the git log.
 """
-Convertion from token assignment to an abstract DB search.
+Conversion from token assignment to an abstract DB search.
 """
 from typing import Optional, List, Tuple, Iterator, Dict
 import heapq
@@ -166,7 +166,7 @@ class SearchBuilder:
         sdata.lookups = [dbf.FieldLookup('name_vector', [t.token for t in hnrs], lookups.LookupAny)]
         expected_count = sum(t.count for t in hnrs)
 
-        partials = {t.token: t.count for trange in address
+        partials = {t.token: t.addr_count for trange in address
                        for t in self.query.get_partials_list(trange)}
 
         if expected_count < 8000:
@@ -176,11 +176,12 @@ class SearchBuilder:
             sdata.lookups.append(dbf.FieldLookup('nameaddress_vector',
                                                  list(partials), lookups.LookupAll))
         else:
+            addr_fulls = [t.token for t
+                          in self.query.get_tokens(address[0], TokenType.WORD)]
+            if len(addr_fulls) > 5:
+                return
             sdata.lookups.append(
-                dbf.FieldLookup('nameaddress_vector',
-                                [t.token for t
-                                 in self.query.get_tokens(address[0], TokenType.WORD)],
-                                lookups.LookupAny))
+                dbf.FieldLookup('nameaddress_vector', addr_fulls, lookups.LookupAny))
 
         sdata.housenumbers = dbf.WeightedStrings([], [])
         yield dbs.PlaceSearch(0.05, sdata, expected_count)
@@ -221,34 +222,85 @@ class SearchBuilder:
             yield penalty, exp_count, dbf.lookup_by_names(list(name_partials.keys()), addr_tokens)
             return
 
+        addr_count = min(t.addr_count for t in addr_partials) if addr_partials else 30000
         # Partial term to frequent. Try looking up by rare full names first.
         name_fulls = self.query.get_tokens(name, TokenType.WORD)
         if name_fulls:
             fulls_count = sum(t.count for t in name_fulls)
-            # At this point drop unindexed partials from the address.
-            # This might yield wrong results, nothing we can do about that.
-            if not partials_indexed:
-                addr_tokens = [t.token for t in addr_partials if t.is_indexed]
+            if len(name_partials) == 1:
+                penalty += min(0.5, max(0, (exp_count - 50 * fulls_count) / (2000 * fulls_count)))
+            if partials_indexed:
                 penalty += 1.2 * sum(t.penalty for t in addr_partials if not t.is_indexed)
-            # Any of the full names applies with all of the partials from the address
-            yield penalty, fulls_count / (2**len(addr_tokens)),\
-                  dbf.lookup_by_any_name([t.token for t in name_fulls],
-                                         addr_tokens,
-                                         fulls_count > 30000 / max(1, len(addr_tokens)))
+
+            if fulls_count < 50000 or addr_count < 30000:
+                yield penalty,fulls_count / (2**len(addr_tokens)), \
+                    self.get_full_name_ranking(name_fulls, addr_partials,
+                                               fulls_count > 30000 / max(1, len(addr_tokens)))
 
         # To catch remaining results, lookup by name and address
         # We only do this if there is a reasonable number of results expected.
         exp_count = exp_count / (2**len(addr_tokens)) if addr_tokens else exp_count
-        if exp_count < 10000 and all(t.is_indexed for t in name_partials.values()):
-            lookup = [dbf.FieldLookup('name_vector', list(name_partials.keys()), lookups.LookupAll)]
-            if addr_tokens:
-                lookup.append(dbf.FieldLookup('nameaddress_vector', addr_tokens, lookups.LookupAll))
+        if exp_count < 10000 and addr_count < 20000\
+           and all(t.is_indexed for t in name_partials.values()):
             penalty += 0.35 * max(1 if name_fulls else 0.1,
                                   5 - len(name_partials) - len(addr_tokens))
-            yield penalty, exp_count, lookup
+            yield penalty, exp_count,\
+                  self.get_name_address_ranking(list(name_partials.keys()), addr_partials)
+
+
+    def get_name_address_ranking(self, name_tokens: List[int],
+                                 addr_partials: List[Token]) -> List[dbf.FieldLookup]:
+        """ Create a ranking expression looking up by name and address.
+        """
+        lookup = [dbf.FieldLookup('name_vector', name_tokens, lookups.LookupAll)]
+
+        addr_restrict_tokens = []
+        addr_lookup_tokens = []
+        for t in addr_partials:
+            if t.is_indexed:
+                if t.addr_count > 20000:
+                    addr_restrict_tokens.append(t.token)
+                else:
+                    addr_lookup_tokens.append(t.token)
+
+        if addr_restrict_tokens:
+            lookup.append(dbf.FieldLookup('nameaddress_vector',
+                                          addr_restrict_tokens, lookups.Restrict))
+        if addr_lookup_tokens:
+            lookup.append(dbf.FieldLookup('nameaddress_vector',
+                                          addr_lookup_tokens, lookups.LookupAll))
+
+        return lookup
+
+
+    def get_full_name_ranking(self, name_fulls: List[Token], addr_partials: List[Token],
+                              use_lookup: bool) -> List[dbf.FieldLookup]:
+        """ Create a ranking expression with full name terms and
+            additional address lookup. When 'use_lookup' is true, then
+            address lookups will use the index, when the occurences are not
+            too many.
+        """
+        # At this point drop unindexed partials from the address.
+        # This might yield wrong results, nothing we can do about that.
+        if use_lookup:
+            addr_restrict_tokens = []
+            addr_lookup_tokens = []
+            for t in addr_partials:
+                if t.is_indexed:
+                    if t.addr_count > 20000:
+                        addr_restrict_tokens.append(t.token)
+                    else:
+                        addr_lookup_tokens.append(t.token)
+        else:
+            addr_restrict_tokens = [t.token for t in addr_partials if t.is_indexed]
+            addr_lookup_tokens = []
 
+        return dbf.lookup_by_any_name([t.token for t in name_fulls],
+                                      addr_restrict_tokens, addr_lookup_tokens)
 
-    def get_name_ranking(self, trange: TokenRange) -> dbf.FieldRanking:
+
+    def get_name_ranking(self, trange: TokenRange,
+                         db_field: str = 'name_vector') -> dbf.FieldRanking:
         """ Create a ranking expression for a name term in the given range.
         """
         name_fulls = self.query.get_tokens(trange, TokenType.WORD)
@@ -257,7 +309,7 @@ class SearchBuilder:
         # Fallback, sum of penalty for partials
         name_partials = self.query.get_partials_list(trange)
         default = sum(t.penalty for t in name_partials) + 0.2
-        return dbf.FieldRanking('name_vector', default, ranks)
+        return dbf.FieldRanking(db_field, default, ranks)
 
 
     def get_addr_ranking(self, trange: TokenRange) -> dbf.FieldRanking:
@@ -315,11 +367,9 @@ class SearchBuilder:
         sdata = dbf.SearchData()
         sdata.penalty = assignment.penalty
         if assignment.country:
-            tokens = self.query.get_tokens(assignment.country, TokenType.COUNTRY)
-            if self.details.countries:
-                tokens = [t for t in tokens if t.lookup_word in self.details.countries]
-                if not tokens:
-                    return None
+            tokens = self.get_country_tokens(assignment.country)
+            if not tokens:
+                return None
             sdata.set_strings('countries', tokens)
         elif self.details.countries:
             sdata.countries = dbf.WeightedStrings(self.details.countries,
@@ -333,24 +383,54 @@ class SearchBuilder:
                               self.query.get_tokens(assignment.postcode,
                                                     TokenType.POSTCODE))
         if assignment.qualifier:
-            tokens = self.query.get_tokens(assignment.qualifier, TokenType.QUALIFIER)
-            if self.details.categories:
-                tokens = [t for t in tokens if t.get_category() in self.details.categories]
-                if not tokens:
-                    return None
+            tokens = self.get_qualifier_tokens(assignment.qualifier)
+            if not tokens:
+                return None
             sdata.set_qualifiers(tokens)
         elif self.details.categories:
             sdata.qualifiers = dbf.WeightedCategories(self.details.categories,
                                                       [0.0] * len(self.details.categories))
 
         if assignment.address:
-            sdata.set_ranking([self.get_addr_ranking(r) for r in assignment.address])
+            if not assignment.name and assignment.housenumber:
+                # housenumber search: the first item needs to be handled like
+                # a name in ranking or penalties are not comparable with
+                # normal searches.
+                sdata.set_ranking([self.get_name_ranking(assignment.address[0],
+                                                         db_field='nameaddress_vector')]
+                                  + [self.get_addr_ranking(r) for r in assignment.address[1:]])
+            else:
+                sdata.set_ranking([self.get_addr_ranking(r) for r in assignment.address])
         else:
             sdata.rankings = []
 
         return sdata
 
 
+    def get_country_tokens(self, trange: TokenRange) -> List[Token]:
+        """ Return the list of country tokens for the given range,
+            optionally filtered by the country list from the details
+            parameters.
+        """
+        tokens = self.query.get_tokens(trange, TokenType.COUNTRY)
+        if self.details.countries:
+            tokens = [t for t in tokens if t.lookup_word in self.details.countries]
+
+        return tokens
+
+
+    def get_qualifier_tokens(self, trange: TokenRange) -> List[Token]:
+        """ Return the list of qualifier tokens for the given range,
+            optionally filtered by the qualifier list from the details
+            parameters.
+        """
+        tokens = self.query.get_tokens(trange, TokenType.QUALIFIER)
+        if self.details.categories:
+            tokens = [t for t in tokens if t.get_category() in self.details.categories]
+
+        return tokens
+
+
     def get_near_items(self, assignment: TokenAssignment) -> Optional[dbf.WeightedCategories]:
         """ Collect tokens for near items search or use the categories
             requested per parameter.