Nominatim/nominatim/tokenizer/icu_tokenizer.py
Sarah Hoffmann d35400a7d7 use analyser provided in the 'analyzer' property
Implements per-name choice of analyzer. If a non-default
analyzer is choosen, then the 'word' identifier is extended
with the name of the ana;yzer, so that we still have unique
items.
2021-10-05 14:10:32 +02:00

647 lines
22 KiB
Python

"""
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 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.base import AbstractAnalyzer, AbstractTokenizer
DBCFG_TERM_NORMALIZATION = "tokenizer_term_normalization"
LOG = logging.getLogger()
def create(dsn, data_dir):
""" Create a new instance of the tokenizer provided by this module.
"""
return LegacyICUTokenizer(dsn, data_dir)
class LegacyICUTokenizer(AbstractTokenizer):
""" This tokenizer uses libICU to covert names and queries to ASCII.
Otherwise it uses the same algorithms and data structures as the
normalization routines in Nominatim 3.
"""
def __init__(self, dsn, data_dir):
self.dsn = dsn
self.data_dir = data_dir
self.loader = None
self.term_normalization = None
def init_new_db(self, config, init_db=True):
""" Set up a new tokenizer for the database.
This copies all necessary data in the project directory to make
sure the tokenizer remains stable even over updates.
"""
self.loader = ICURuleLoader(config)
self.term_normalization = config.TERM_NORMALIZATION
self._install_php(config.lib_dir.php)
self._save_config()
if init_db:
self.update_sql_functions(config)
self._init_db_tables(config)
def init_from_project(self, config):
""" Initialise the tokenizer from the project directory.
"""
self.loader = ICURuleLoader(config)
with connect(self.dsn) as conn:
self.loader.load_config_from_db(conn)
self.term_normalization = get_property(conn, DBCFG_TERM_NORMALIZATION)
def finalize_import(self, _):
""" Do any required postprocessing to make the tokenizer data ready
for use.
"""
def update_sql_functions(self, config):
""" Reimport the SQL functions for this tokenizer.
"""
with connect(self.dsn) as conn:
sqlp = SQLPreprocessor(conn, config)
sqlp.run_sql_file(conn, 'tokenizer/icu_tokenizer.sql')
def check_database(self, config):
""" Check that the tokenizer is set up correctly.
"""
self.init_from_project(config)
if self.term_normalization is None:
return "Configuration for tokenizer 'icu' are missing."
return None
def name_analyzer(self):
""" Create a new analyzer for tokenizing names and queries
using this tokinzer. Analyzers are context managers and should
be used accordingly:
```
with tokenizer.name_analyzer() as analyzer:
analyser.tokenize()
```
When used outside the with construct, the caller must ensure to
call the close() function before destructing the analyzer.
Analyzers are not thread-safe. You need to instantiate one per thread.
"""
return LegacyICUNameAnalyzer(self.dsn, self.loader.make_sanitizer(),
self.loader.make_token_analysis())
def _install_php(self, phpdir):
""" Install the php script for the tokenizer.
"""
php_file = self.data_dir / "tokenizer.php"
php_file.write_text(dedent(f"""\
<?php
@define('CONST_Max_Word_Frequency', 10000000);
@define('CONST_Term_Normalization_Rules', "{self.term_normalization}");
@define('CONST_Transliteration', "{self.loader.get_search_rules()}");
require_once('{phpdir}/tokenizer/icu_tokenizer.php');"""))
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.loader.save_config_to_db(conn)
set_property(conn, DBCFG_TERM_NORMALIZATION, self.term_normalization)
def _init_db_tables(self, config):
""" Set up the word table and fill it with pre-computed word
frequencies.
"""
with connect(self.dsn) as conn:
sqlp = SQLPreprocessor(conn, config)
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()
analysis = self.loader.make_token_analysis()
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:
word = analysis.search.transliterate(name)
if word and ' ' in word:
for term in set(word.split()):
words[term] += cnt
return words
class LegacyICUNameAnalyzer(AbstractAnalyzer):
""" The legacy analyzer uses the ICU library for splitting names.
Each instance opens a connection to the database to request the
normalization.
"""
def __init__(self, dsn, sanitizer, token_analysis):
self.conn = connect(dsn).connection
self.conn.autocommit = True
self.sanitizer = sanitizer
self.token_analysis = token_analysis
self._cache = _TokenCache()
def close(self):
""" Free all resources used by the analyzer.
"""
if self.conn:
self.conn.close()
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
otherwise is a partial name.
The function returns a list of tuples with
(original word, word token, word id).
The function is used for testing and debugging only
and not necessarily efficient.
"""
full_tokens = {}
partial_tokens = {}
for word in words:
if word.startswith('#'):
full_tokens[word] = self._search_normalized(word[1:])
else:
partial_tokens[word] = self._search_normalized(word)
with self.conn.cursor() as cur:
cur.execute("""SELECT word_token, word_id
FROM word WHERE word_token = ANY(%s) and type = 'W'
""", (list(full_tokens.values()),))
full_ids = {r[0]: r[1] for r in cur}
cur.execute("""SELECT word_token, word_id
FROM word WHERE word_token = ANY(%s) and type = 'w'""",
(list(partial_tokens.values()),))
part_ids = {r[0]: r[1] for r in cur}
return [(k, v, full_ids.get(v, None)) for k, v in full_tokens.items()] \
+ [(k, v, part_ids.get(v, None)) for k, v in partial_tokens.items()]
@staticmethod
def normalize_postcode(postcode):
""" Convert the postcode to a standardized form.
This function must yield exactly the same result as the SQL function
'token_normalized_postcode()'.
"""
return postcode.strip().upper()
def _make_standard_hnr(self, hnr):
""" Create a normalised version of a housenumber.
This function takes minor shortcuts on transliteration.
"""
return self._search_normalized(hnr)
def update_postcodes_from_db(self):
""" Update postcode tokens in the word table from the location_postcode
table.
"""
to_delete = []
with self.conn.cursor() as cur:
# This finds us the rows in location_postcode and word that are
# missing in the other table.
cur.execute("""SELECT * FROM
(SELECT pc, word FROM
(SELECT distinct(postcode) as pc FROM location_postcode) p
FULL JOIN
(SELECT word FROM word WHERE type = 'P') w
ON pc = word) x
WHERE pc is null or word is null""")
with CopyBuffer() as copystr:
for postcode, word in cur:
if postcode is None:
to_delete.append(word)
else:
copystr.add(self._search_normalized(postcode),
'P', postcode)
if to_delete:
cur.execute("""DELETE FROM WORD
WHERE type ='P' and word = any(%s)
""", (to_delete, ))
copystr.copy_out(cur, 'word',
columns=['word_token', 'type', 'word'])
def update_special_phrases(self, phrases, should_replace):
""" Replace the search index for special phrases with the new phrases.
If `should_replace` is True, then the previous set of will be
completely replaced. Otherwise the phrases are added to the
already existing ones.
"""
norm_phrases = set(((self._normalized(p[0]), p[1], p[2], p[3])
for p in phrases))
with self.conn.cursor() as cur:
# Get the old phrases.
existing_phrases = set()
cur.execute("SELECT word, info FROM word WHERE type = 'S'")
for word, info in cur:
existing_phrases.add((word, info['class'], info['type'],
info.get('op') or '-'))
added = self._add_special_phrases(cur, norm_phrases, existing_phrases)
if should_replace:
deleted = self._remove_special_phrases(cur, norm_phrases,
existing_phrases)
else:
deleted = 0
LOG.info("Total phrases: %s. Added: %s. Deleted: %s",
len(norm_phrases), added, deleted)
def _add_special_phrases(self, cursor, new_phrases, existing_phrases):
""" Add all phrases to the database that are not yet there.
"""
to_add = new_phrases - existing_phrases
added = 0
with CopyBuffer() as copystr:
for word, cls, typ, oper in to_add:
term = self._search_normalized(word)
if term:
copystr.add(term, 'S', word,
json.dumps({'class': cls, 'type': typ,
'op': oper if oper in ('in', 'near') else None}))
added += 1
copystr.copy_out(cursor, 'word',
columns=['word_token', 'type', 'word', 'info'])
return added
@staticmethod
def _remove_special_phrases(cursor, new_phrases, existing_phrases):
""" Remove all phrases from the databse that are no longer in the
new phrase list.
"""
to_delete = existing_phrases - new_phrases
if to_delete:
cursor.execute_values(
""" DELETE FROM word USING (VALUES %s) as v(name, in_class, in_type, op)
WHERE type = 'S' and word = name
and info->>'class' = in_class and info->>'type' = in_type
and ((op = '-' and info->>'op' is null) or op = info->>'op')
""", to_delete)
return len(to_delete)
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 names:
norm_name = self._search_normalized(name.name)
if norm_name:
word_tokens.add(norm_name)
with self.conn.cursor() as cur:
# Get existing names
cur.execute("""SELECT word_token FROM word
WHERE type = 'C' and word = %s""",
(country_code, ))
word_tokens.difference_update((t[0] for t in cur))
# Only add those names that are not yet in the list.
if word_tokens:
cur.execute("""INSERT INTO word (word_token, type, word)
(SELECT token, 'C', %s
FROM unnest(%s) as token)
""", (country_code, list(word_tokens)))
# No names are deleted at the moment.
# If deletion is made possible, then the static names from the
# initial 'country_name' table should be kept.
def process_place(self, place):
""" Determine tokenizer information about the given place.
Returns a JSON-serializable structure that will be handed into
the database via the token_info field.
"""
token_info = _TokenInfo(self._cache)
names, address = self.sanitizer.process_names(place)
if names:
fulls, partials = self._compute_name_tokens(names)
token_info.add_names(fulls, partials)
if place.is_country():
self._add_country_full_names(place.country_code, names)
if address:
self._process_place_address(token_info, address)
return token_info.data
def _process_place_address(self, token_info, address):
hnrs = []
addr_terms = []
for item in address:
if item.kind == 'postcode':
self._add_postcode(item.name)
elif item.kind in ('housenumber', 'streetnumber', 'conscriptionnumber'):
hnrs.append(item.name)
elif item.kind == 'street':
token_info.add_street(self._compute_partial_tokens(item.name))
elif item.kind == 'place':
token_info.add_place(self._compute_partial_tokens(item.name))
elif not item.kind.startswith('_') 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])
if addr_terms:
token_info.add_address_terms(addr_terms)
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 _compute_name_tokens(self, names):
""" Computes the full name and partial name tokens for the given
dictionary of names.
"""
full_tokens = set()
partial_tokens = set()
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.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)).*",
(token_id, variants))
full, part = cur.fetchone()
self._cache.names[token_id] = (full, part)
full_tokens.add(full)
partial_tokens.update(part)
return full_tokens, partial_tokens
def _add_postcode(self, postcode):
""" Make sure the normalized postcode is present in the word table.
"""
if re.search(r'[:,;]', postcode) is None:
postcode = self.normalize_postcode(postcode)
if postcode not in self._cache.postcodes:
term = self._search_normalized(postcode)
if not term:
return
with self.conn.cursor() as cur:
# no word_id needed for postcodes
cur.execute("""INSERT INTO word (word_token, type, word)
(SELECT %s, 'P', pc FROM (VALUES (%s)) as v(pc)
WHERE NOT EXISTS
(SELECT * FROM word
WHERE type = 'P' and word = pc))
""", (term, postcode))
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.
"""
def __init__(self, cache):
self._cache = cache
self.data = {}
@staticmethod
def _mk_array(tokens):
return '{%s}' % ','.join((str(s) for s in tokens))
def add_names(self, fulls, partials):
""" Adds token information for the normalised names.
"""
self.data['names'] = self._mk_array(itertools.chain(fulls, partials))
def add_housenumbers(self, conn, hnrs):
""" Extract housenumber information from a list of normalised
housenumbers.
"""
self.data['hnr_tokens'] = self._mk_array(self._cache.get_hnr_tokens(conn, hnrs))
self.data['hnr'] = ';'.join(hnrs)
def add_street(self, tokens):
""" Add addr:street match terms.
"""
if tokens:
self.data['street'] = self._mk_array(tokens)
def add_place(self, tokens):
""" Add addr:place search and match terms.
"""
if tokens:
self.data['place'] = self._mk_array(tokens)
def add_address_terms(self, terms):
""" Add additional address terms.
"""
tokens = {key: self._mk_array(partials)
for key, partials in terms if partials}
if tokens:
self.data['addr'] = tokens
class _TokenCache:
""" Cache for token information to avoid repeated database queries.
This cache is not thread-safe and needs to be instantiated per
analyzer.
"""
def __init__(self):
self.names = {}
self.partials = {}
self.postcodes = set()
self.housenumbers = {}
def get_hnr_tokens(self, conn, terms):
""" Get token ids for a list of housenumbers, looking them up in the
database if necessary. `terms` is an iterable of normalized
housenumbers.
"""
tokens = []
askdb = []
for term in terms:
token = self.housenumbers.get(term)
if token is None:
askdb.append(term)
else:
tokens.append(token)
if askdb:
with conn.cursor() as cur:
cur.execute("SELECT nr, getorcreate_hnr_id(nr) FROM unnest(%s) as nr",
(askdb, ))
for term, tid in cur:
self.housenumbers[term] = tid
tokens.append(tid)
return tokens