X-Git-Url: http://git.veekun.com/zzz-pokedex.git/blobdiff_plain/706d0f3ea899d26403b412545d60877996e81994..734bf27fd89587d1feda1dc61e0ad5e294c24094:/pokedex/lookup.py diff --git a/pokedex/lookup.py b/pokedex/lookup.py index 115e9f0..8c3e444 100644 --- a/pokedex/lookup.py +++ b/pokedex/lookup.py @@ -1,7 +1,5 @@ # encoding: utf8 -from collections import namedtuple import os, os.path -import pkg_resources import random import re import shutil @@ -16,9 +14,12 @@ from whoosh.qparser import QueryParser import whoosh.scoring import whoosh.spelling +from pokedex.util import namedtuple + from pokedex.db import connect import pokedex.db.tables as tables from pokedex.roomaji import romanize +from pokedex.defaults import get_default_index_dir __all__ = ['PokedexLookup'] @@ -28,28 +29,66 @@ rx_is_number = re.compile('^\d+$') LookupResult = namedtuple('LookupResult', ['object', 'indexed_name', 'name', 'language', 'iso3166', 'exact']) +class UninitializedIndex(object): + class UninitializedIndexError(Exception): + pass + + def __nonzero__(self): + """Dummy object should identify itself as False.""" + return False + + def __bool__(self): + """Python 3000 version of the above. Future-proofing rules!""" + return False + + def __getattr__(self, *args, **kwargs): + raise self.UninitializedIndexError( + "The lookup index does not exist. Please use `pokedex setup` " + "or lookup.rebuild_index() to create it." + ) + class LanguageWeighting(whoosh.scoring.Weighting): """A scoring class that forces otherwise-equal English results to come before foreign results. """ + def __init__(self, extra_weights={}, *args, **kwargs): + """`extra_weights` may be a dictionary of weights which will be + factored in. + + Intended for use with spelling corrections, which come along with their + own weightings. + """ + self.extra_weights = extra_weights + super(LanguageWeighting, self).__init__(*args, **kwargs) + def score(self, searcher, fieldnum, text, docnum, weight, QTF=1): doc = searcher.stored_fields(docnum) + + # Apply extra weight + weight = weight * self.extra_weights.get(text, 1.0) + if doc['language'] == None: # English (well, "default"); leave it at 1 return weight elif doc['language'] == u'Roomaji': # Give Roomaji a little boost; it's most likely to be searched - return weight * 0.95 + return weight * 0.9 else: # Everything else can drop down the totem pole - return weight * 0.9 + return weight * 0.8 class PokedexLookup(object): INTERMEDIATE_LOOKUP_RESULTS = 25 MAX_LOOKUP_RESULTS = 10 + # The speller only checks how much the input matches a word; there can be + # all manner of extra unmatched junk, and it won't affect the weighting. + # To compensate, greatly boost the weighting of matches at the beginning + # and end, so nearly-full-word-matches are much better + SPELLER_OPTIONS = dict(booststart=10.0, boostend=9.0) + # Dictionary of table name => table class. # Need the table name so we can get the class from the table name after we # retrieve something from the index @@ -67,7 +106,7 @@ class PokedexLookup(object): ) - def __init__(self, directory=None, session=None, recreate=False): + def __init__(self, directory=None, session=None): """Opens the whoosh index stored in the named directory. If the index doesn't already exist, it will be created. @@ -76,22 +115,19 @@ class PokedexLookup(object): `pokedex` egg directory. `session` - If the index needs to be created, this database session will be - used. Defaults to an attempt to connect to the default SQLite - database installed by `pokedex setup`. - - `recreate` - If set to True, the whoosh index will be created even if it already - exists. + Used for creating the index and retrieving objects. Defaults to an + attempt to connect to the default SQLite database installed by + `pokedex setup`. """ # By the time this returns, self.index, self.speller, and self.session # must be set - # Defaults - if not directory: - directory = pkg_resources.resource_filename('pokedex', - 'data/whoosh-index') + # If a directory was not given, use the default + if directory is None: + directory = get_default_index_dir() + + self.directory = directory if session: self.session = session @@ -99,31 +135,34 @@ class PokedexLookup(object): self.session = connect() # Attempt to open or create the index - directory_exists = os.path.exists(directory) - if directory_exists and not recreate: - # Already exists; should be an index! Bam, done. - try: - self.index = whoosh.index.open_dir(directory, indexname='MAIN') - spell_store = whoosh.filedb.filestore.FileStorage(directory) - self.speller = whoosh.spelling.SpellChecker(spell_store) - return - except whoosh.index.EmptyIndexError as e: - # Apparently not a real index. Fall out and create it - pass - - # Delete and start over if we're going to bail anyway. - if directory_exists and recreate: - # Be safe and only delete if it looks like a whoosh index, i.e., - # everything starts with _ - if all(f[0] == '_' for f in os.listdir(directory)): - shutil.rmtree(directory) - directory_exists = False - - if not directory_exists: - os.mkdir(directory) - - - ### Create index + if not os.path.exists(directory) or not os.listdir(directory): + # Directory doesn't exist OR is empty; caller needs to use + # rebuild_index before doing anything. Provide a dummy object that + # complains when used + self.index = UninitializedIndex() + self.speller = UninitializedIndex() + return + + # Otherwise, already exists; should be an index! Bam, done. + # Note that this will explode if the directory exists but doesn't + # contain an index; that's a feature + try: + self.index = whoosh.index.open_dir(directory, indexname='MAIN') + except whoosh.index.EmptyIndexError: + raise IOError( + "The index directory already contains files. " + "Please use a dedicated directory for the lookup index." + ) + + # Create speller, and done + spell_store = whoosh.filedb.filestore.FileStorage(directory) + self.speller = whoosh.spelling.SpellChecker(spell_store, + **self.SPELLER_OPTIONS) + + + def rebuild_index(self): + """Creates the index from scratch.""" + schema = whoosh.fields.Schema( name=whoosh.fields.ID(stored=True), table=whoosh.fields.ID(stored=True), @@ -133,16 +172,15 @@ class PokedexLookup(object): display_name=whoosh.fields.STORED, # non-lowercased name ) - self.index = whoosh.index.create_in(directory, schema=schema, - indexname='MAIN') + if not os.path.exists(self.directory): + os.mkdir(self.directory) + + self.index = whoosh.index.create_in(self.directory, schema=schema, + indexname='MAIN') writer = self.index.writer() # Index every name in all our tables of interest - # speller_entries becomes a list of (word, score) tuples; the score is - # 2 for English names, 1.5 for Roomaji, and 1 for everything else. I - # think this biases the results in the direction most people expect, - # especially when e.g. German names are very similar to English names - speller_entries = [] + speller_entries = set() for cls in self.indexed_tables.values(): q = self.session.query(cls) @@ -150,7 +188,7 @@ class PokedexLookup(object): row_key = dict(table=unicode(cls.__tablename__), row_id=unicode(row.id)) - def add(name, language, iso3166, score): + def add(name, language, iso3166): normalized_name = self.normalize_name(name) writer.add_document( @@ -159,21 +197,21 @@ class PokedexLookup(object): **row_key ) - speller_entries.append((normalized_name, score)) + speller_entries.add(normalized_name) # Add the basic English name to the index if cls == tables.Pokemon: # Pokémon need their form name added # XXX kinda kludgy - add(row.full_name, None, u'us', 1) + add(row.full_name, None, u'us') # If this is a default form, ALSO add the unadorned name, # so 'Deoxys' alone will still do the right thing if row.forme_name and not row.forme_base_pokemon_id: - add(row.name, None, u'us', 1) + add(row.name, None, u'us') else: - add(row.name, None, u'us', 1) + add(row.name, None, u'us') # Some things also have other languages' names # XXX other language form names..? @@ -186,21 +224,21 @@ class PokedexLookup(object): continue add(moonspeak, foreign_name.language.name, - foreign_name.language.iso3166, - 3) + foreign_name.language.iso3166) # Add Roomaji too if foreign_name.language.name == 'Japanese': roomaji = romanize(foreign_name.name) - add(roomaji, u'Roomaji', u'jp', 8) + add(roomaji, u'Roomaji', u'jp') writer.commit() # Construct and populate a spell-checker index. Quicker to do it all # at once, as every call to add_* does a commit(), and those seem to be # expensive - self.speller = whoosh.spelling.SpellChecker(self.index.storage) - self.speller.add_scored_words(speller_entries) + self.speller = whoosh.spelling.SpellChecker(self.index.storage, mingram=2, + **self.SPELLER_OPTIONS) + self.speller.add_words(speller_entries) def normalize_name(self, name): @@ -290,15 +328,17 @@ class PokedexLookup(object): Returns None for a bogus name. """ + # Table object if hasattr(name, '__tablename__'): return getattr(name, '__tablename__') - elif name in self.indexed_tables: - return name - elif name + 's' in self.indexed_tables: - return name + 's' - else: - # Bogus. Be nice and return dummy - return None + + # Table name + for table in self.indexed_tables.values(): + if name in (table.__tablename__, table.__singlename__): + return table.__tablename__ + + # Bogus. Be nice and return dummy + return None def _whoosh_records_to_results(self, records, exact=True): """Converts a list of whoosh's indexed records to LookupResult tuples @@ -387,13 +427,20 @@ class PokedexLookup(object): # Do different things depending what the query looks like # Note: Term objects do an exact match, so we don't have to worry about # a query parser tripping on weird characters in the input + try: + # Let Python try to convert to a number, so 0xff works + name_as_number = int(name, base=0) + except ValueError: + # Oh well + name_as_number = None + if '*' in name or '?' in name: exact_only = True query = whoosh.query.Wildcard(u'name', name) - elif rx_is_number.match(name): + elif name_as_number is not None: # Don't spell-check numbers! exact_only = True - query = whoosh.query.Term(u'row_id', name) + query = whoosh.query.Term(u'row_id', unicode(name_as_number)) else: # Not an integer query = whoosh.query.Term(u'name', name) @@ -415,11 +462,30 @@ class PokedexLookup(object): exact = False results = [] - for suggestion in self.speller.suggest( - name, self.INTERMEDIATE_LOOKUP_RESULTS): - - query = whoosh.query.Term('name', suggestion) - results.extend(searcher.search(query)) + fuzzy_query_parts = [] + fuzzy_weights = {} + min_weight = [None] + for suggestion, _, weight in self.speller.suggestions_and_scores(name): + # Only allow the top 50% of scores; otherwise there will always + # be a lot of trailing junk + if min_weight[0] is None: + min_weight[0] = weight * 0.5 + elif weight < min_weight[0]: + break + + fuzzy_query_parts.append(whoosh.query.Term('name', suggestion)) + fuzzy_weights[suggestion] = weight + + if not fuzzy_query_parts: + # Nothing at all; don't try querying + return [] + + fuzzy_query = whoosh.query.Or(fuzzy_query_parts) + if type_term: + fuzzy_query = fuzzy_query & type_term + + searcher.weighting = LanguageWeighting(extra_weights=fuzzy_weights) + results = searcher.search(fuzzy_query) ### Convert results to db objects objects = self._whoosh_records_to_results(results, exact=exact)