1 # SPDX-License-Identifier: GPL-3.0-or-later
3 # This file is part of Nominatim. (https://nominatim.org)
5 # Copyright (C) 2024 by the Nominatim developer community.
6 # For a full list of authors see the git log.
8 Conversion from token assignment to an abstract DB search.
10 from typing import Optional, List, Tuple, Iterator, Dict
13 from ..types import SearchDetails, DataLayer
14 from .query import QueryStruct, Token, TokenType, TokenRange, BreakType
15 from .token_assignment import TokenAssignment
16 from . import db_search_fields as dbf
17 from . import db_searches as dbs
18 from . import db_search_lookups as lookups
21 def wrap_near_search(categories: List[Tuple[str, str]],
22 search: dbs.AbstractSearch) -> dbs.NearSearch:
23 """ Create a new search that wraps the given search in a search
24 for near places of the given category.
26 return dbs.NearSearch(penalty=search.penalty,
27 categories=dbf.WeightedCategories(categories,
28 [0.0] * len(categories)),
32 def build_poi_search(category: List[Tuple[str, str]],
33 countries: Optional[List[str]]) -> dbs.PoiSearch:
34 """ Create a new search for places by the given category, possibly
35 constraint to the given countries.
38 ccs = dbf.WeightedStrings(countries, [0.0] * len(countries))
40 ccs = dbf.WeightedStrings([], [])
42 class _PoiData(dbf.SearchData):
44 qualifiers = dbf.WeightedCategories(category, [0.0] * len(category))
47 return dbs.PoiSearch(_PoiData())
51 """ Build the abstract search queries from token assignments.
54 def __init__(self, query: QueryStruct, details: SearchDetails) -> None:
56 self.details = details
59 def configured_for_country(self) -> bool:
60 """ Return true if the search details are configured to
61 allow countries in the result.
63 return self.details.min_rank <= 4 and self.details.max_rank >= 4 \
64 and self.details.layer_enabled(DataLayer.ADDRESS)
67 def configured_for_postcode(self) -> bool:
68 """ Return true if the search details are configured to
69 allow postcodes in the result.
71 return self.details.min_rank <= 5 and self.details.max_rank >= 11\
72 and self.details.layer_enabled(DataLayer.ADDRESS)
75 def configured_for_housenumbers(self) -> bool:
76 """ Return true if the search details are configured to
77 allow addresses in the result.
79 return self.details.max_rank >= 30 \
80 and self.details.layer_enabled(DataLayer.ADDRESS)
82 def build(self, assignment: TokenAssignment) -> Iterator[dbs.AbstractSearch]:
83 """ Yield all possible abstract searches for the given token assignment.
85 sdata = self.get_search_data(assignment)
89 near_items = self.get_near_items(assignment)
90 if near_items is not None and not near_items:
91 return # impossible combination of near items and category parameter
93 if assignment.name is None:
94 if near_items and not sdata.postcodes:
95 sdata.qualifiers = near_items
97 builder = self.build_poi_search(sdata)
98 elif assignment.housenumber:
99 hnr_tokens = self.query.get_tokens(assignment.housenumber,
100 TokenType.HOUSENUMBER)
101 builder = self.build_housenumber_search(sdata, hnr_tokens, assignment.address)
103 builder = self.build_special_search(sdata, assignment.address,
106 builder = self.build_name_search(sdata, assignment.name, assignment.address,
110 penalty = min(near_items.penalties)
111 near_items.penalties = [p - penalty for p in near_items.penalties]
112 for search in builder:
113 search_penalty = search.penalty
115 yield dbs.NearSearch(penalty + assignment.penalty + search_penalty,
118 for search in builder:
119 search.penalty += assignment.penalty
122 def build_poi_search(self, sdata: dbf.SearchData) -> Iterator[dbs.AbstractSearch]:
123 """ Build abstract search query for a simple category search.
124 This kind of search requires an additional geographic constraint.
126 if not sdata.housenumbers \
127 and ((self.details.viewbox and self.details.bounded_viewbox) or self.details.near):
128 yield dbs.PoiSearch(sdata)
130 def build_special_search(self, sdata: dbf.SearchData,
131 address: List[TokenRange],
132 is_category: bool) -> Iterator[dbs.AbstractSearch]:
133 """ Build abstract search queries for searches that do not involve
137 # No special searches over qualifiers supported.
140 if sdata.countries and not address and not sdata.postcodes \
141 and self.configured_for_country:
142 yield dbs.CountrySearch(sdata)
144 if sdata.postcodes and (is_category or self.configured_for_postcode):
145 penalty = 0.0 if sdata.countries else 0.1
147 sdata.lookups = [dbf.FieldLookup('nameaddress_vector',
148 [t.token for r in address
149 for t in self.query.get_partials_list(r)],
152 yield dbs.PostcodeSearch(penalty, sdata)
154 def build_housenumber_search(self, sdata: dbf.SearchData, hnrs: List[Token],
155 address: List[TokenRange]) -> Iterator[dbs.AbstractSearch]:
156 """ Build a simple address search for special entries where the
157 housenumber is the main name token.
159 sdata.lookups = [dbf.FieldLookup('name_vector', [t.token for t in hnrs], lookups.LookupAny)]
160 expected_count = sum(t.count for t in hnrs)
162 partials = {t.token: t.addr_count for trange in address
163 for t in self.query.get_partials_list(trange)}
166 # can happen when none of the partials is indexed
169 if expected_count < 8000:
170 sdata.lookups.append(dbf.FieldLookup('nameaddress_vector',
171 list(partials), lookups.Restrict))
172 elif len(partials) != 1 or list(partials.values())[0] < 10000:
173 sdata.lookups.append(dbf.FieldLookup('nameaddress_vector',
174 list(partials), lookups.LookupAll))
176 addr_fulls = [t.token for t
177 in self.query.get_tokens(address[0], TokenType.WORD)]
178 if len(addr_fulls) > 5:
180 sdata.lookups.append(
181 dbf.FieldLookup('nameaddress_vector', addr_fulls, lookups.LookupAny))
183 sdata.housenumbers = dbf.WeightedStrings([], [])
184 yield dbs.PlaceSearch(0.05, sdata, expected_count)
186 def build_name_search(self, sdata: dbf.SearchData,
187 name: TokenRange, address: List[TokenRange],
188 is_category: bool) -> Iterator[dbs.AbstractSearch]:
189 """ Build abstract search queries for simple name or address searches.
191 if is_category or not sdata.housenumbers or self.configured_for_housenumbers:
192 ranking = self.get_name_ranking(name)
193 name_penalty = ranking.normalize_penalty()
195 sdata.rankings.append(ranking)
196 for penalty, count, lookup in self.yield_lookups(name, address):
197 sdata.lookups = lookup
198 yield dbs.PlaceSearch(penalty + name_penalty, sdata, count)
200 def yield_lookups(self, name: TokenRange, address: List[TokenRange]
201 ) -> Iterator[Tuple[float, int, List[dbf.FieldLookup]]]:
202 """ Yield all variants how the given name and address should best
203 be searched for. This takes into account how frequent the terms
204 are and tries to find a lookup that optimizes index use.
206 penalty = 0.0 # extra penalty
207 name_partials = {t.token: t for t in self.query.get_partials_list(name)}
209 addr_partials = [t for r in address for t in self.query.get_partials_list(r)]
210 addr_tokens = list({t.token for t in addr_partials})
212 exp_count = min(t.count for t in name_partials.values()) / (2**(len(name_partials) - 1))
214 if (len(name_partials) > 3 or exp_count < 8000):
215 yield penalty, exp_count, dbf.lookup_by_names(list(name_partials.keys()), addr_tokens)
218 addr_count = min(t.addr_count for t in addr_partials) if addr_partials else 50000
219 # Partial term to frequent. Try looking up by rare full names first.
220 name_fulls = self.query.get_tokens(name, TokenType.WORD)
222 fulls_count = sum(t.count for t in name_fulls)
224 if fulls_count < 80000 or addr_count < 50000:
225 yield penalty, fulls_count / (2**len(addr_tokens)), \
226 self.get_full_name_ranking(name_fulls, addr_partials,
227 fulls_count > 30000 / max(1, len(addr_tokens)))
229 # To catch remaining results, lookup by name and address
230 # We only do this if there is a reasonable number of results expected.
231 exp_count = exp_count / (2**len(addr_tokens)) if addr_tokens else exp_count
232 if exp_count < 10000 and addr_count < 20000:
233 penalty += 0.35 * max(1 if name_fulls else 0.1,
234 5 - len(name_partials) - len(addr_tokens))
235 yield penalty, exp_count, \
236 self.get_name_address_ranking(list(name_partials.keys()), addr_partials)
238 def get_name_address_ranking(self, name_tokens: List[int],
239 addr_partials: List[Token]) -> List[dbf.FieldLookup]:
240 """ Create a ranking expression looking up by name and address.
242 lookup = [dbf.FieldLookup('name_vector', name_tokens, lookups.LookupAll)]
244 addr_restrict_tokens = []
245 addr_lookup_tokens = []
246 for t in addr_partials:
247 if t.addr_count > 20000:
248 addr_restrict_tokens.append(t.token)
250 addr_lookup_tokens.append(t.token)
252 if addr_restrict_tokens:
253 lookup.append(dbf.FieldLookup('nameaddress_vector',
254 addr_restrict_tokens, lookups.Restrict))
255 if addr_lookup_tokens:
256 lookup.append(dbf.FieldLookup('nameaddress_vector',
257 addr_lookup_tokens, lookups.LookupAll))
261 def get_full_name_ranking(self, name_fulls: List[Token], addr_partials: List[Token],
262 use_lookup: bool) -> List[dbf.FieldLookup]:
263 """ Create a ranking expression with full name terms and
264 additional address lookup. When 'use_lookup' is true, then
265 address lookups will use the index, when the occurrences are not
268 # At this point drop unindexed partials from the address.
269 # This might yield wrong results, nothing we can do about that.
271 addr_restrict_tokens = []
272 addr_lookup_tokens = [t.token for t in addr_partials]
274 addr_restrict_tokens = [t.token for t in addr_partials]
275 addr_lookup_tokens = []
277 return dbf.lookup_by_any_name([t.token for t in name_fulls],
278 addr_restrict_tokens, addr_lookup_tokens)
280 def get_name_ranking(self, trange: TokenRange,
281 db_field: str = 'name_vector') -> dbf.FieldRanking:
282 """ Create a ranking expression for a name term in the given range.
284 name_fulls = self.query.get_tokens(trange, TokenType.WORD)
285 ranks = [dbf.RankedTokens(t.penalty, [t.token]) for t in name_fulls]
286 ranks.sort(key=lambda r: r.penalty)
287 # Fallback, sum of penalty for partials
288 name_partials = self.query.get_partials_list(trange)
289 default = sum(t.penalty for t in name_partials) + 0.2
290 return dbf.FieldRanking(db_field, default, ranks)
292 def get_addr_ranking(self, trange: TokenRange) -> dbf.FieldRanking:
293 """ Create a list of ranking expressions for an address term
294 for the given ranges.
296 todo: List[Tuple[int, int, dbf.RankedTokens]] = []
297 heapq.heappush(todo, (0, trange.start, dbf.RankedTokens(0.0, [])))
298 ranks: List[dbf.RankedTokens] = []
301 neglen, pos, rank = heapq.heappop(todo)
302 for tlist in self.query.nodes[pos].starting:
303 if tlist.ttype in (TokenType.PARTIAL, TokenType.WORD):
304 if tlist.end < trange.end:
305 chgpenalty = PENALTY_WORDCHANGE[self.query.nodes[tlist.end].btype]
306 if tlist.ttype == TokenType.PARTIAL:
307 penalty = rank.penalty + chgpenalty \
308 + max(t.penalty for t in tlist.tokens)
309 heapq.heappush(todo, (neglen - 1, tlist.end,
310 dbf.RankedTokens(penalty, rank.tokens)))
312 for t in tlist.tokens:
313 heapq.heappush(todo, (neglen - 1, tlist.end,
314 rank.with_token(t, chgpenalty)))
315 elif tlist.end == trange.end:
316 if tlist.ttype == TokenType.PARTIAL:
317 ranks.append(dbf.RankedTokens(rank.penalty
318 + max(t.penalty for t in tlist.tokens),
321 ranks.extend(rank.with_token(t, 0.0) for t in tlist.tokens)
323 # Too many variants, bail out and only add
324 # Worst-case Fallback: sum of penalty of partials
325 name_partials = self.query.get_partials_list(trange)
326 default = sum(t.penalty for t in name_partials) + 0.2
327 ranks.append(dbf.RankedTokens(rank.penalty + default, []))
328 # Bail out of outer loop
332 ranks.sort(key=lambda r: len(r.tokens))
333 default = ranks[0].penalty + 0.3
335 ranks.sort(key=lambda r: r.penalty)
337 return dbf.FieldRanking('nameaddress_vector', default, ranks)
339 def get_search_data(self, assignment: TokenAssignment) -> Optional[dbf.SearchData]:
340 """ Collect the tokens for the non-name search fields in the
343 sdata = dbf.SearchData()
344 sdata.penalty = assignment.penalty
345 if assignment.country:
346 tokens = self.get_country_tokens(assignment.country)
349 sdata.set_strings('countries', tokens)
350 elif self.details.countries:
351 sdata.countries = dbf.WeightedStrings(self.details.countries,
352 [0.0] * len(self.details.countries))
353 if assignment.housenumber:
354 sdata.set_strings('housenumbers',
355 self.query.get_tokens(assignment.housenumber,
356 TokenType.HOUSENUMBER))
357 if assignment.postcode:
358 sdata.set_strings('postcodes',
359 self.query.get_tokens(assignment.postcode,
361 if assignment.qualifier:
362 tokens = self.get_qualifier_tokens(assignment.qualifier)
365 sdata.set_qualifiers(tokens)
366 elif self.details.categories:
367 sdata.qualifiers = dbf.WeightedCategories(self.details.categories,
368 [0.0] * len(self.details.categories))
370 if assignment.address:
371 if not assignment.name and assignment.housenumber:
372 # housenumber search: the first item needs to be handled like
373 # a name in ranking or penalties are not comparable with
375 sdata.set_ranking([self.get_name_ranking(assignment.address[0],
376 db_field='nameaddress_vector')]
377 + [self.get_addr_ranking(r) for r in assignment.address[1:]])
379 sdata.set_ranking([self.get_addr_ranking(r) for r in assignment.address])
385 def get_country_tokens(self, trange: TokenRange) -> List[Token]:
386 """ Return the list of country tokens for the given range,
387 optionally filtered by the country list from the details
390 tokens = self.query.get_tokens(trange, TokenType.COUNTRY)
391 if self.details.countries:
392 tokens = [t for t in tokens if t.lookup_word in self.details.countries]
396 def get_qualifier_tokens(self, trange: TokenRange) -> List[Token]:
397 """ Return the list of qualifier tokens for the given range,
398 optionally filtered by the qualifier list from the details
401 tokens = self.query.get_tokens(trange, TokenType.QUALIFIER)
402 if self.details.categories:
403 tokens = [t for t in tokens if t.get_category() in self.details.categories]
407 def get_near_items(self, assignment: TokenAssignment) -> Optional[dbf.WeightedCategories]:
408 """ Collect tokens for near items search or use the categories
409 requested per parameter.
410 Returns None if no category search is requested.
412 if assignment.near_item:
413 tokens: Dict[Tuple[str, str], float] = {}
414 for t in self.query.get_tokens(assignment.near_item, TokenType.NEAR_ITEM):
415 cat = t.get_category()
416 # The category of a near search will be that of near_item.
417 # Thus, if search is restricted to a category parameter,
418 # the two sets must intersect.
419 if (not self.details.categories or cat in self.details.categories)\
420 and t.penalty < tokens.get(cat, 1000.0):
421 tokens[cat] = t.penalty
422 return dbf.WeightedCategories(list(tokens.keys()), list(tokens.values()))
427 PENALTY_WORDCHANGE = {
428 BreakType.START: 0.0,
430 BreakType.PHRASE: 0.0,