X-Git-Url: https://git.openstreetmap.org./nominatim.git/blobdiff_plain/7f9cb4e68d14b0460982bee64af7b2467c2ecd1e..1b7c8240baeb383a2481f895fc85a74aaa16e94e:/nominatim/api/search/token_assignment.py?ds=inline diff --git a/nominatim/api/search/token_assignment.py b/nominatim/api/search/token_assignment.py index 33fb7335..d94d6903 100644 --- 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 - category: Optional[qmod.TokenRange] = None + near_item: 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 - 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 @@ -109,7 +109,7 @@ class _TokenSequence: """ # 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]: @@ -165,22 +165,22 @@ class _TokenSequence: 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 - 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 - 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 - 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: @@ -253,6 +253,8 @@ 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 + if any(t.ttype == qmod.TokenType.NEAR_ITEM for t in self.seq): + self.penalty += 1.0 return True @@ -288,18 +290,29 @@ class _TokenSequence: yield dataclasses.replace(base, penalty=self.penalty, name=first, address=base.address[1:]) - if (not base.housenumber or first.end >= base.housenumber.start)\ - and (not base.qualifier or first.start >= base.qualifier.end): - base_penalty = self.penalty - if (base.housenumber and base.housenumber.start > first.start) \ - or len(query.source) > 1: - base_penalty += 0.25 - for i in range(first.start + 1, first.end): - name, addr = first.split(i) - penalty = base_penalty + PENALTY_TOKENCHANGE[query.nodes[i].btype] - log().comment(f'split first word = name ({i - first.start})') - yield dataclasses.replace(base, name=name, penalty=penalty, - address=[addr] + base.address[1:]) + # To paraphrase: + # * if another name term comes after the first one and before the + # housenumber + # * a qualifier comes after the name + # * the containing phrase is strictly typed + if (base.housenumber and first.end < base.housenumber.start)\ + or (base.qualifier and base.qualifier > first)\ + or (query.nodes[first.start].ptype != qmod.PhraseType.NONE): + return + + penalty = self.penalty + + # Penalty for: + # * , , , ... + # * queries that are comma-separated + if (base.housenumber and base.housenumber > first) or len(query.source) > 1: + penalty += 0.25 + + for i in range(first.start + 1, first.end): + name, addr = first.split(i) + log().comment(f'split first word = name ({i - first.start})') + yield dataclasses.replace(base, name=name, address=[addr] + base.address[1:], + penalty=penalty + PENALTY_TOKENCHANGE[query.nodes[i].btype]) def _get_assignments_address_backward(self, base: TokenAssignment, @@ -314,19 +327,27 @@ class _TokenSequence: yield dataclasses.replace(base, penalty=self.penalty, name=last, address=base.address[:-1]) - if (not base.housenumber or last.start <= base.housenumber.end)\ - and (not base.qualifier or last.end <= base.qualifier.start): - base_penalty = self.penalty - if base.housenumber and base.housenumber.start < last.start: - base_penalty += 0.4 - if len(query.source) > 1: - base_penalty += 0.25 - for i in range(last.start + 1, last.end): - addr, name = last.split(i) - penalty = base_penalty + PENALTY_TOKENCHANGE[query.nodes[i].btype] - log().comment(f'split last word = name ({i - last.start})') - yield dataclasses.replace(base, name=name, penalty=penalty, - address=base.address[:-1] + [addr]) + # To paraphrase: + # * if another name term comes before the last one and after the + # housenumber + # * a qualifier comes before the name + # * the containing phrase is strictly typed + if (base.housenumber and last.start > base.housenumber.end)\ + or (base.qualifier and base.qualifier < last)\ + or (query.nodes[last.start].ptype != qmod.PhraseType.NONE): + return + + penalty = self.penalty + if base.housenumber and base.housenumber < last: + penalty += 0.4 + if len(query.source) > 1: + penalty += 0.25 + + for i in range(last.start + 1, last.end): + addr, name = last.split(i) + log().comment(f'split last word = name ({i - last.start})') + yield dataclasses.replace(base, name=name, address=base.address[:-1] + [addr], + penalty=penalty + PENALTY_TOKENCHANGE[query.nodes[i].btype]) def get_assignments(self, query: qmod.QueryStruct) -> Iterator[TokenAssignment]: @@ -347,7 +368,7 @@ class _TokenSequence: # 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: