X-Git-Url: https://git.openstreetmap.org./nominatim.git/blobdiff_plain/925195725dfcb7f1a6795c50244c1df6cb7242ce..d8b7a51ab6d47da79c7e281465c9f5a7bfd74a5a:/nominatim/tokenizer/icu_tokenizer.py

diff --git a/nominatim/tokenizer/icu_tokenizer.py b/nominatim/tokenizer/icu_tokenizer.py
index cb411204..0841300a 100644
--- a/nominatim/tokenizer/icu_tokenizer.py
+++ b/nominatim/tokenizer/icu_tokenizer.py
@@ -1,24 +1,26 @@
+# SPDX-License-Identifier: GPL-2.0-only
+#
+# This file is part of Nominatim. (https://nominatim.org)
+#
+# Copyright (C) 2022 by the Nominatim developer community.
+# For a full list of authors see the git log.
 """
 Tokenizer implementing normalisation as used before Nominatim 4 but using
 libICU instead of the PostgreSQL module.
 """
-from collections import Counter
 import itertools
 import json
 import logging
 import re
 from textwrap import dedent
-from pathlib import Path
 
 from nominatim.db.connection import connect
-from nominatim.db.properties import set_property, get_property
 from nominatim.db.utils import CopyBuffer
 from nominatim.db.sql_preprocessor import SQLPreprocessor
+from nominatim.indexer.place_info import PlaceInfo
 from nominatim.tokenizer.icu_rule_loader import ICURuleLoader
-from nominatim.tokenizer.icu_name_processor import ICUNameProcessor, ICUNameProcessorRules
 from nominatim.tokenizer.base import AbstractAnalyzer, AbstractTokenizer
 
-DBCFG_MAXWORDFREQ = "tokenizer_maxwordfreq"
 DBCFG_TERM_NORMALIZATION = "tokenizer_term_normalization"
 
 LOG = logging.getLogger()
@@ -38,9 +40,7 @@ class LegacyICUTokenizer(AbstractTokenizer):
     def __init__(self, dsn, data_dir):
         self.dsn = dsn
         self.data_dir = data_dir
-        self.naming_rules = None
-        self.term_normalization = None
-        self.max_word_frequency = None
+        self.loader = None
 
 
     def init_new_db(self, config, init_db=True):
@@ -49,58 +49,106 @@ class LegacyICUTokenizer(AbstractTokenizer):
             This copies all necessary data in the project directory to make
             sure the tokenizer remains stable even over updates.
         """
-        if config.TOKENIZER_CONFIG:
-            cfgfile = Path(config.TOKENIZER_CONFIG)
-        else:
-            cfgfile = config.config_dir / 'icu_tokenizer.yaml'
-
-        loader = ICURuleLoader(cfgfile)
-        self.naming_rules = ICUNameProcessorRules(loader=loader)
-        self.term_normalization = config.TERM_NORMALIZATION
-        self.max_word_frequency = config.MAX_WORD_FREQUENCY
+        self.loader = ICURuleLoader(config)
 
         self._install_php(config.lib_dir.php)
-        self._save_config(config)
+        self._save_config()
 
         if init_db:
             self.update_sql_functions(config)
             self._init_db_tables(config)
 
 
-    def init_from_project(self):
+    def init_from_project(self, config):
         """ Initialise the tokenizer from the project directory.
         """
+        self.loader = ICURuleLoader(config)
+
         with connect(self.dsn) as conn:
-            self.naming_rules = ICUNameProcessorRules(conn=conn)
-            self.term_normalization = get_property(conn, DBCFG_TERM_NORMALIZATION)
-            self.max_word_frequency = get_property(conn, DBCFG_MAXWORDFREQ)
+            self.loader.load_config_from_db(conn)
 
 
-    def finalize_import(self, _):
+    def finalize_import(self, config):
         """ Do any required postprocessing to make the tokenizer data ready
             for use.
         """
+        with connect(self.dsn) as conn:
+            sqlp = SQLPreprocessor(conn, config)
+            sqlp.run_sql_file(conn, 'tokenizer/legacy_tokenizer_indices.sql')
 
 
     def update_sql_functions(self, config):
         """ Reimport the SQL functions for this tokenizer.
         """
         with connect(self.dsn) as conn:
-            max_word_freq = get_property(conn, DBCFG_MAXWORDFREQ)
             sqlp = SQLPreprocessor(conn, config)
-            sqlp.run_sql_file(conn, 'tokenizer/icu_tokenizer.sql',
-                              max_word_freq=max_word_freq)
+            sqlp.run_sql_file(conn, 'tokenizer/icu_tokenizer.sql')
 
 
-    def check_database(self):
+    def check_database(self, config):
         """ Check that the tokenizer is set up correctly.
         """
-        self.init_from_project()
+        # Will throw an error if there is an issue.
+        self.init_from_project(config)
+
+
+    def update_statistics(self):
+        """ Recompute frequencies for all name words.
+        """
+        with connect(self.dsn) as conn:
+            if conn.table_exists('search_name'):
+                with conn.cursor() as cur:
+                    cur.drop_table("word_frequencies")
+                    LOG.info("Computing word frequencies")
+                    cur.execute("""CREATE TEMP TABLE word_frequencies AS
+                                     SELECT unnest(name_vector) as id, count(*)
+                                     FROM search_name GROUP BY id""")
+                    cur.execute("CREATE INDEX ON word_frequencies(id)")
+                    LOG.info("Update word table with recomputed frequencies")
+                    cur.execute("""UPDATE word
+                                   SET info = info || jsonb_build_object('count', count)
+                                   FROM word_frequencies WHERE word_id = id""")
+                    cur.drop_table("word_frequencies")
+            conn.commit()
+
+
+    def _cleanup_housenumbers(self):
+        """ Remove unused house numbers.
+        """
+        with connect(self.dsn) as conn:
+            with conn.cursor(name="hnr_counter") as cur:
+                cur.execute("""SELECT word_id, word_token FROM word
+                               WHERE type = 'H'
+                                 AND NOT EXISTS(SELECT * FROM search_name
+                                                WHERE ARRAY[word.word_id] && name_vector)
+                                 AND (char_length(word_token) > 6
+                                      OR word_token not similar to '\d+')
+                            """)
+                candidates = {token: wid for wid, token in cur}
+            with conn.cursor(name="hnr_counter") as cur:
+                cur.execute("""SELECT housenumber FROM placex
+                               WHERE housenumber is not null
+                                     AND (char_length(housenumber) > 6
+                                          OR housenumber not similar to '\d+')
+                            """)
+                for row in cur:
+                    for hnr in row[0].split(';'):
+                        candidates.pop(hnr, None)
+            LOG.info("There are %s outdated housenumbers.", len(candidates))
+            if candidates:
+                with conn.cursor() as cur:
+                    cur.execute("""DELETE FROM word WHERE word_id = any(%s)""",
+                                (list(candidates.values()), ))
+                conn.commit()
 
-        if self.naming_rules is None:
-            return "Configuration for tokenizer 'icu' are missing."
 
-        return None
+
+    def update_word_tokens(self):
+        """ Remove unused tokens.
+        """
+        LOG.warn("Cleaning up housenumber tokens.")
+        self._cleanup_housenumbers()
+        LOG.warn("Tokenizer house-keeping done.")
 
 
     def name_analyzer(self):
@@ -118,7 +166,8 @@ class LegacyICUTokenizer(AbstractTokenizer):
 
             Analyzers are not thread-safe. You need to instantiate one per thread.
         """
-        return LegacyICUNameAnalyzer(self.dsn, ICUNameProcessor(self.naming_rules))
+        return LegacyICUNameAnalyzer(self.dsn, self.loader.make_sanitizer(),
+                                     self.loader.make_token_analysis())
 
 
     def _install_php(self, phpdir):
@@ -127,21 +176,18 @@ class LegacyICUTokenizer(AbstractTokenizer):
         php_file = self.data_dir / "tokenizer.php"
         php_file.write_text(dedent(f"""\
             <?php
-            @define('CONST_Max_Word_Frequency', {self.max_word_frequency});
-            @define('CONST_Term_Normalization_Rules', "{self.term_normalization}");
-            @define('CONST_Transliteration', "{self.naming_rules.search_rules}");
+            @define('CONST_Max_Word_Frequency', 10000000);
+            @define('CONST_Term_Normalization_Rules', "{self.loader.normalization_rules}");
+            @define('CONST_Transliteration', "{self.loader.get_search_rules()}");
             require_once('{phpdir}/tokenizer/icu_tokenizer.php');"""))
 
 
-    def _save_config(self, config):
+    def _save_config(self):
         """ Save the configuration that needs to remain stable for the given
             database as database properties.
         """
         with connect(self.dsn) as conn:
-            self.naming_rules.save_rules(conn)
-
-            set_property(conn, DBCFG_MAXWORDFREQ, config.MAX_WORD_FREQUENCY)
-            set_property(conn, DBCFG_TERM_NORMALIZATION, self.term_normalization)
+            self.loader.save_config_to_db(conn)
 
 
     def _init_db_tables(self, config):
@@ -153,45 +199,6 @@ class LegacyICUTokenizer(AbstractTokenizer):
             sqlp.run_sql_file(conn, 'tokenizer/icu_tokenizer_tables.sql')
             conn.commit()
 
-            LOG.warning("Precomputing word tokens")
-
-            # get partial words and their frequencies
-            words = self._count_partial_terms(conn)
-
-            # copy them back into the word table
-            with CopyBuffer() as copystr:
-                for term, cnt in words.items():
-                    copystr.add('w', term, json.dumps({'count': cnt}))
-
-                with conn.cursor() as cur:
-                    copystr.copy_out(cur, 'word',
-                                     columns=['type', 'word_token', 'info'])
-                    cur.execute("""UPDATE word SET word_id = nextval('seq_word')
-                                   WHERE word_id is null and type = 'w'""")
-
-            conn.commit()
-
-    def _count_partial_terms(self, conn):
-        """ Count the partial terms from the names in the place table.
-        """
-        words = Counter()
-        name_proc = ICUNameProcessor(self.naming_rules)
-
-        with conn.cursor(name="words") as cur:
-            cur.execute(""" SELECT v, count(*) FROM
-                              (SELECT svals(name) as v FROM place)x
-                            WHERE length(v) < 75 GROUP BY v""")
-
-            for name, cnt in cur:
-                terms = set()
-                for word in name_proc.get_variants_ascii(name_proc.get_normalized(name)):
-                    if ' ' in word:
-                        terms.update(word.split())
-                for term in terms:
-                    words[term] += cnt
-
-        return words
-
 
 class LegacyICUNameAnalyzer(AbstractAnalyzer):
     """ The legacy analyzer uses the ICU library for splitting names.
@@ -200,10 +207,11 @@ class LegacyICUNameAnalyzer(AbstractAnalyzer):
         normalization.
     """
 
-    def __init__(self, dsn, name_proc):
+    def __init__(self, dsn, sanitizer, token_analysis):
         self.conn = connect(dsn).connection
         self.conn.autocommit = True
-        self.name_processor = name_proc
+        self.sanitizer = sanitizer
+        self.token_analysis = token_analysis
 
         self._cache = _TokenCache()
 
@@ -216,6 +224,19 @@ class LegacyICUNameAnalyzer(AbstractAnalyzer):
             self.conn = None
 
 
+    def _search_normalized(self, name):
+        """ Return the search token transliteration of the given name.
+        """
+        return self.token_analysis.search.transliterate(name).strip()
+
+
+    def _normalized(self, name):
+        """ Return the normalized version of the given name with all
+            non-relevant information removed.
+        """
+        return self.token_analysis.normalizer.transliterate(name).strip()
+
+
     def get_word_token_info(self, words):
         """ Return token information for the given list of words.
             If a word starts with # it is assumed to be a full name
@@ -231,9 +252,9 @@ class LegacyICUNameAnalyzer(AbstractAnalyzer):
         partial_tokens = {}
         for word in words:
             if word.startswith('#'):
-                full_tokens[word] = self.name_processor.get_search_normalized(word[1:])
+                full_tokens[word] = self._search_normalized(word[1:])
             else:
-                partial_tokens[word] = self.name_processor.get_search_normalized(word)
+                partial_tokens[word] = self._search_normalized(word)
 
         with self.conn.cursor() as cur:
             cur.execute("""SELECT word_token, word_id
@@ -264,7 +285,7 @@ class LegacyICUNameAnalyzer(AbstractAnalyzer):
 
             This function takes minor shortcuts on transliteration.
         """
-        return self.name_processor.get_search_normalized(hnr)
+        return self._search_normalized(hnr)
 
     def update_postcodes_from_db(self):
         """ Update postcode tokens in the word table from the location_postcode
@@ -287,7 +308,7 @@ class LegacyICUNameAnalyzer(AbstractAnalyzer):
                     if postcode is None:
                         to_delete.append(word)
                     else:
-                        copystr.add(self.name_processor.get_search_normalized(postcode),
+                        copystr.add(self._search_normalized(postcode),
                                     'P', postcode)
 
                 if to_delete:
@@ -305,7 +326,7 @@ class LegacyICUNameAnalyzer(AbstractAnalyzer):
             completely replaced. Otherwise the phrases are added to the
             already existing ones.
         """
-        norm_phrases = set(((self.name_processor.get_normalized(p[0]), p[1], p[2], p[3])
+        norm_phrases = set(((self._normalized(p[0]), p[1], p[2], p[3])
                             for p in phrases))
 
         with self.conn.cursor() as cur:
@@ -335,7 +356,7 @@ class LegacyICUNameAnalyzer(AbstractAnalyzer):
         added = 0
         with CopyBuffer() as copystr:
             for word, cls, typ, oper in to_add:
-                term = self.name_processor.get_search_normalized(word)
+                term = self._search_normalized(word)
                 if term:
                     copystr.add(term, 'S', word,
                                 json.dumps({'class': cls, 'type': typ,
@@ -369,9 +390,21 @@ class LegacyICUNameAnalyzer(AbstractAnalyzer):
     def add_country_names(self, country_code, names):
         """ Add names for the given country to the search index.
         """
+        # Make sure any name preprocessing for country names applies.
+        info = PlaceInfo({'name': names, 'country_code': country_code,
+                          'rank_address': 4, 'class': 'boundary',
+                          'type': 'administrative'})
+        self._add_country_full_names(country_code,
+                                     self.sanitizer.process_names(info)[0])
+
+
+    def _add_country_full_names(self, country_code, names):
+        """ Add names for the given country from an already sanitized
+            name list.
+        """
         word_tokens = set()
-        for name in self._compute_full_names(names):
-            norm_name = self.name_processor.get_search_normalized(name)
+        for name in names:
+            norm_name = self._search_normalized(name.name)
             if norm_name:
                 word_tokens.add(norm_name)
 
@@ -397,23 +430,21 @@ class LegacyICUNameAnalyzer(AbstractAnalyzer):
     def process_place(self, place):
         """ Determine tokenizer information about the given place.
 
-            Returns a JSON-serialisable structure that will be handed into
+            Returns a JSON-serializable structure that will be handed into
             the database via the token_info field.
         """
         token_info = _TokenInfo(self._cache)
 
-        names = place.get('name')
+        names, address = self.sanitizer.process_names(place)
 
         if names:
             fulls, partials = self._compute_name_tokens(names)
 
             token_info.add_names(fulls, partials)
 
-            country_feature = place.get('country_feature')
-            if country_feature and re.fullmatch(r'[A-Za-z][A-Za-z]', country_feature):
-                self.add_country_names(country_feature.lower(), names)
+            if place.is_country():
+                self._add_country_full_names(place.country_code, names)
 
-        address = place.get('address')
         if address:
             self._process_place_address(token_info, address)
 
@@ -421,51 +452,110 @@ class LegacyICUNameAnalyzer(AbstractAnalyzer):
 
 
     def _process_place_address(self, token_info, address):
-        hnrs = []
+        hnrs = set()
         addr_terms = []
-        for key, value in address.items():
-            if key == 'postcode':
-                self._add_postcode(value)
-            elif key in ('housenumber', 'streetnumber', 'conscriptionnumber'):
-                hnrs.append(value)
-            elif key == 'street':
-                token_info.add_street(*self._compute_name_tokens({'name': value}))
-            elif key == 'place':
-                token_info.add_place(*self._compute_name_tokens({'name': value}))
-            elif not key.startswith('_') and \
-                 key not in ('country', 'full'):
-                addr_terms.append((key, *self._compute_name_tokens({'name': value})))
+        streets = []
+        for item in address:
+            if item.kind == 'postcode':
+                self._add_postcode(item.name)
+            elif item.kind == 'housenumber':
+                norm_name = self._make_standard_hnr(item.name)
+                if norm_name:
+                    hnrs.add(norm_name)
+            elif item.kind == 'street':
+                streets.extend(self._retrieve_full_tokens(item.name))
+            elif item.kind == 'place':
+                if not item.suffix:
+                    token_info.add_place(self._compute_partial_tokens(item.name))
+            elif not item.kind.startswith('_') and not item.suffix and \
+                 item.kind not in ('country', 'full'):
+                addr_terms.append((item.kind, self._compute_partial_tokens(item.name)))
 
         if hnrs:
-            hnrs = self._split_housenumbers(hnrs)
-            token_info.add_housenumbers(self.conn, [self._make_standard_hnr(n) for n in hnrs])
+            token_info.add_housenumbers(self.conn, hnrs)
 
         if addr_terms:
             token_info.add_address_terms(addr_terms)
 
+        if streets:
+            token_info.add_street(streets)
+
+
+    def _compute_partial_tokens(self, name):
+        """ Normalize the given term, split it into partial words and return
+            then token list for them.
+        """
+        norm_name = self._search_normalized(name)
+
+        tokens = []
+        need_lookup = []
+        for partial in norm_name.split():
+            token = self._cache.partials.get(partial)
+            if token:
+                tokens.append(token)
+            else:
+                need_lookup.append(partial)
+
+        if need_lookup:
+            with self.conn.cursor() as cur:
+                cur.execute("""SELECT word, getorcreate_partial_word(word)
+                               FROM unnest(%s) word""",
+                            (need_lookup, ))
+
+                for partial, token in cur:
+                    tokens.append(token)
+                    self._cache.partials[partial] = token
+
+        return tokens
+
+
+    def _retrieve_full_tokens(self, name):
+        """ Get the full name token for the given name, if it exists.
+            The name is only retrived for the standard analyser.
+        """
+        norm_name = self._search_normalized(name)
+
+        # return cached if possible
+        if norm_name in self._cache.fulls:
+            return self._cache.fulls[norm_name]
+
+        with self.conn.cursor() as cur:
+            cur.execute("SELECT word_id FROM word WHERE word_token = %s and type = 'W'",
+                        (norm_name, ))
+            full = [row[0] for row in cur]
+
+        self._cache.fulls[norm_name] = full
+
+        return full
+
 
     def _compute_name_tokens(self, names):
         """ Computes the full name and partial name tokens for the given
             dictionary of names.
         """
-        full_names = self._compute_full_names(names)
         full_tokens = set()
         partial_tokens = set()
 
-        for name in full_names:
-            norm_name = self.name_processor.get_normalized(name)
-            full, part = self._cache.names.get(norm_name, (None, None))
+        for name in names:
+            analyzer_id = name.get_attr('analyzer')
+            norm_name = self._normalized(name.name)
+            if analyzer_id is None:
+                token_id = norm_name
+            else:
+                token_id = f'{norm_name}@{analyzer_id}'
+
+            full, part = self._cache.names.get(token_id, (None, None))
             if full is None:
-                variants = self.name_processor.get_variants_ascii(norm_name)
+                variants = self.token_analysis.analysis[analyzer_id].get_variants_ascii(norm_name)
                 if not variants:
                     continue
 
                 with self.conn.cursor() as cur:
                     cur.execute("SELECT (getorcreate_full_word(%s, %s)).*",
-                                (norm_name, variants))
+                                (token_id, variants))
                     full, part = cur.fetchone()
 
-                self._cache.names[norm_name] = (full, part)
+                self._cache.names[token_id] = (full, part)
 
             full_tokens.add(full)
             partial_tokens.update(part)
@@ -473,23 +563,6 @@ class LegacyICUNameAnalyzer(AbstractAnalyzer):
         return full_tokens, partial_tokens
 
 
-    @staticmethod
-    def _compute_full_names(names):
-        """ Return the set of all full name word ids to be used with the
-            given dictionary of names.
-        """
-        full_names = set()
-        for name in (n.strip() for ns in names.values() for n in re.split('[;,]', ns)):
-            if name:
-                full_names.add(name)
-
-                brace_idx = name.find('(')
-                if brace_idx >= 0:
-                    full_names.add(name[:brace_idx].strip())
-
-        return full_names
-
-
     def _add_postcode(self, postcode):
         """ Make sure the normalized postcode is present in the word table.
         """
@@ -497,7 +570,7 @@ class LegacyICUNameAnalyzer(AbstractAnalyzer):
             postcode = self.normalize_postcode(postcode)
 
             if postcode not in self._cache.postcodes:
-                term = self.name_processor.get_search_normalized(postcode)
+                term = self._search_normalized(postcode)
                 if not term:
                     return
 
@@ -512,24 +585,6 @@ class LegacyICUNameAnalyzer(AbstractAnalyzer):
                 self._cache.postcodes.add(postcode)
 
 
-    @staticmethod
-    def _split_housenumbers(hnrs):
-        if len(hnrs) > 1 or ',' in hnrs[0] or ';' in hnrs[0]:
-            # split numbers if necessary
-            simple_list = []
-            for hnr in hnrs:
-                simple_list.extend((x.strip() for x in re.split(r'[;,]', hnr)))
-
-            if len(simple_list) > 1:
-                hnrs = list(set(simple_list))
-            else:
-                hnrs = simple_list
-
-        return hnrs
-
-
-
-
 class _TokenInfo:
     """ Collect token information to be sent back to the database.
     """
@@ -556,30 +611,24 @@ class _TokenInfo:
         self.data['hnr'] = ';'.join(hnrs)
 
 
-    def add_street(self, fulls, _):
+    def add_street(self, tokens):
         """ Add addr:street match terms.
         """
-        if fulls:
-            self.data['street'] = self._mk_array(fulls)
+        self.data['street'] = self._mk_array(tokens)
 
 
-    def add_place(self, fulls, partials):
+    def add_place(self, tokens):
         """ Add addr:place search and match terms.
         """
-        if fulls:
-            self.data['place_search'] = self._mk_array(itertools.chain(fulls, partials))
-            self.data['place_match'] = self._mk_array(fulls)
+        if tokens:
+            self.data['place'] = self._mk_array(tokens)
 
 
     def add_address_terms(self, terms):
         """ Add additional address terms.
         """
-        tokens = {}
-
-        for key, fulls, partials in terms:
-            if fulls:
-                tokens[key] = [self._mk_array(itertools.chain(fulls, partials)),
-                               self._mk_array(fulls)]
+        tokens = {key: self._mk_array(partials)
+                  for key, partials in terms if partials}
 
         if tokens:
             self.data['addr'] = tokens
@@ -593,6 +642,8 @@ class _TokenCache:
     """
     def __init__(self):
         self.names = {}
+        self.partials = {}
+        self.fulls = {}
         self.postcodes = set()
         self.housenumbers = {}