Merge pull request #3346 from lonvia/reduce-artificial-importance

[nominatim.git] / nominatim / api / search / token_assignment.py

diff --git a/nominatim/api/search/token_assignment.py b/nominatim/api/search/token_assignment.py
index 3f0e737b004d623f5eef758a40b1a59e1d495806..7a53a20efe35bd80c3f43cc68aeab5e27894983b 100644 (file)
--- a/nominatim/api/search/token_assignment.py
+++ b/nominatim/api/search/token_assignment.py
@@ -46,7 +46,7 @@ class TokenAssignment: # pylint: disable=too-many-instance-attributes
     housenumber: Optional[qmod.TokenRange] = None
     postcode: Optional[qmod.TokenRange] = None
     country: Optional[qmod.TokenRange] = None
     housenumber: Optional[qmod.TokenRange] = None
     postcode: Optional[qmod.TokenRange] = None
     country: Optional[qmod.TokenRange] = None

-    category: Optional[qmod.TokenRange] = None
+    near_item: Optional[qmod.TokenRange] = None

     qualifier: Optional[qmod.TokenRange] = None
 
 
     qualifier: Optional[qmod.TokenRange] = None
 
 


@@ -64,8 +64,8 @@ class TokenAssignment: # pylint: disable=too-many-instance-attributes
                 out.postcode = token.trange
             elif token.ttype == qmod.TokenType.COUNTRY:
                 out.country = token.trange
                 out.postcode = token.trange
             elif token.ttype == qmod.TokenType.COUNTRY:
                 out.country = token.trange

-            elif token.ttype == qmod.TokenType.CATEGORY:
-                out.category = token.trange
+            elif token.ttype == qmod.TokenType.NEAR_ITEM:
+                out.near_item = token.trange

             elif token.ttype == qmod.TokenType.QUALIFIER:
                 out.qualifier = token.trange
         return out
             elif token.ttype == qmod.TokenType.QUALIFIER:
                 out.qualifier = token.trange
         return out


@@ -109,7 +109,7 @@ class _TokenSequence:
         """
         # Country and category must be the final term for left-to-right
         return len(self.seq) > 1 and \
         """
         # Country and category must be the final term for left-to-right
         return len(self.seq) > 1 and \

-               self.seq[-1].ttype in (qmod.TokenType.COUNTRY, qmod.TokenType.CATEGORY)
+               self.seq[-1].ttype in (qmod.TokenType.COUNTRY, qmod.TokenType.NEAR_ITEM)

 
 
     def appendable(self, ttype: qmod.TokenType) -> Optional[int]:
 
 
     def appendable(self, ttype: qmod.TokenType) -> Optional[int]:


@@ -132,6 +132,11 @@ class _TokenSequence:
 
         # Name tokens are always acceptable and don't change direction
         if ttype == qmod.TokenType.PARTIAL:
 
         # Name tokens are always acceptable and don't change direction
         if ttype == qmod.TokenType.PARTIAL:

+            # qualifiers cannot appear in the middle of the qeury. They need
+            # to be near the next phrase.
+            if self.direction == -1 \
+               and any(t.ttype == qmod.TokenType.QUALIFIER for t in self.seq[:-1]):
+                return None

             return self.direction
 
         # Other tokens may only appear once
             return self.direction
 
         # Other tokens may only appear once


@@ -165,22 +170,22 @@ class _TokenSequence:
         if ttype == qmod.TokenType.COUNTRY:
             return None if self.direction == -1 else 1
 
         if ttype == qmod.TokenType.COUNTRY:
             return None if self.direction == -1 else 1
 

-        if ttype == qmod.TokenType.CATEGORY:
+        if ttype == qmod.TokenType.NEAR_ITEM:

             return self.direction
 
         if ttype == qmod.TokenType.QUALIFIER:
             if self.direction == 1:
                 if (len(self.seq) == 1
             return self.direction
 
         if ttype == qmod.TokenType.QUALIFIER:
             if self.direction == 1:
                 if (len(self.seq) == 1

-                    and self.seq[0].ttype in (qmod.TokenType.PARTIAL, qmod.TokenType.CATEGORY)) \
+                    and self.seq[0].ttype in (qmod.TokenType.PARTIAL, qmod.TokenType.NEAR_ITEM)) \

                    or (len(self.seq) == 2
                    or (len(self.seq) == 2

-                       and self.seq[0].ttype == qmod.TokenType.CATEGORY
+                       and self.seq[0].ttype == qmod.TokenType.NEAR_ITEM

                        and self.seq[1].ttype == qmod.TokenType.PARTIAL):
                     return 1
                 return None
             if self.direction == -1:
                 return -1
 
                        and self.seq[1].ttype == qmod.TokenType.PARTIAL):
                     return 1
                 return None
             if self.direction == -1:
                 return -1
 

-            tempseq = self.seq[1:] if self.seq[0].ttype == qmod.TokenType.CATEGORY else self.seq
+            tempseq = self.seq[1:] if self.seq[0].ttype == qmod.TokenType.NEAR_ITEM else self.seq

             if len(tempseq) == 0:
                 return 1
             if len(tempseq) == 1 and self.seq[0].ttype == qmod.TokenType.HOUSENUMBER:
             if len(tempseq) == 0:
                 return 1
             if len(tempseq) == 1 and self.seq[0].ttype == qmod.TokenType.HOUSENUMBER:


@@ -253,7 +258,7 @@ class _TokenSequence:
                 priors = sum(1 for t in self.seq[hnrpos+1:] if t.ttype == qmod.TokenType.PARTIAL)
                 if not self._adapt_penalty_from_priors(priors, 1):
                     return False
                 priors = sum(1 for t in self.seq[hnrpos+1:] if t.ttype == qmod.TokenType.PARTIAL)
                 if not self._adapt_penalty_from_priors(priors, 1):
                     return False

-            if any(t.ttype == qmod.TokenType.CATEGORY for t in self.seq):
+            if any(t.ttype == qmod.TokenType.NEAR_ITEM for t in self.seq):

                 self.penalty += 1.0
 
         return True
                 self.penalty += 1.0
 
         return True


@@ -368,7 +373,7 @@ class _TokenSequence:
 
         # Postcode or country-only search
         if not base.address:
 
         # Postcode or country-only search
         if not base.address:

-            if not base.housenumber and (base.postcode or base.country or base.category):
+            if not base.housenumber and (base.postcode or base.country or base.near_item):

                 log().comment('postcode/country search')
                 yield dataclasses.replace(base, penalty=self.penalty)
         else:
                 log().comment('postcode/country search')
                 yield dataclasses.replace(base, penalty=self.penalty)
         else:


@@ -385,7 +390,7 @@ class _TokenSequence:
                 yield from self._get_assignments_address_backward(base, query)
 
             # variant for special housenumber searches
                 yield from self._get_assignments_address_backward(base, query)
 
             # variant for special housenumber searches

-            if base.housenumber:
+            if base.housenumber and not base.qualifier:

                 yield dataclasses.replace(base, penalty=self.penalty)
 
 
                 yield dataclasses.replace(base, penalty=self.penalty)