X-Git-Url: http://git.veekun.com/zzz-pokedex.git/blobdiff_plain/b7268b5c97e6c0f5c2b92e5befb53d1a251e42b3..929cfc5f8a786c162a0cbdbb30153fe8d1984d7b:/pokedex/lookup.py diff --git a/pokedex/lookup.py b/pokedex/lookup.py index 55e1b5a..0fa91cb 100644 --- a/pokedex/lookup.py +++ b/pokedex/lookup.py @@ -1,4 +1,5 @@ # encoding: utf8 +from collections import namedtuple import os, os.path import pkg_resources import re @@ -9,10 +10,14 @@ import whoosh.filedb.filestore import whoosh.filedb.fileindex import whoosh.index from whoosh.qparser import QueryParser +import whoosh.scoring import whoosh.spelling from pokedex.db import connect import pokedex.db.tables as tables +from pokedex.roomaji import romanize + +__all__ = ['open_index', 'lookup'] # Dictionary of table name => table class. # Need the table name so we can get the class from the table name after we @@ -27,17 +32,6 @@ for cls in [ ]: indexed_tables[cls.__tablename__] = cls -# Dictionary of extra keys to file types of objects under, e.g. Pokémon can -# also be looked up purely by number -extra_keys = { - tables.Move: [ - lambda row: u"move %d" % row.id, - ], - tables.Pokemon: [ - lambda row: unicode(row.id), - ], -} - def open_index(directory=None, session=None, recreate=False): """Opens the whoosh index stored in the named directory and returns (index, speller). If the index doesn't already exist, it will be created. @@ -69,8 +63,9 @@ def open_index(directory=None, session=None, recreate=False): if directory_exists and not recreate: # Already exists; should be an index! try: - index = whoosh.index.open_dir(directory, indexname='pokedex') - speller = whoosh.index.open_dir(directory, indexname='spelling') + index = whoosh.index.open_dir(directory, indexname='MAIN') + spell_store = whoosh.filedb.filestore.FileStorage(directory) + speller = whoosh.spelling.SpellChecker(spell_store) return index, speller except whoosh.index.EmptyIndexError as e: # Apparently not a real index. Fall out of the if and create it @@ -84,19 +79,18 @@ def open_index(directory=None, session=None, recreate=False): schema = whoosh.fields.Schema( name=whoosh.fields.ID(stored=True), table=whoosh.fields.STORED, - row_id=whoosh.fields.STORED, + row_id=whoosh.fields.ID(stored=True), language=whoosh.fields.STORED, - - # Whoosh 0.2 explodes when using a file-stored schema with no TEXT - # columns. Appease it - dummy=whoosh.fields.TEXT, ) - index = whoosh.index.create_in(directory, schema=schema, - indexname='pokedex') + index = whoosh.index.create_in(directory, schema=schema, indexname='MAIN') writer = 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 = [] for cls in indexed_tables.values(): q = session.query(cls) @@ -106,74 +100,86 @@ def open_index(directory=None, session=None, recreate=False): q = q.filter_by(forme_base_pokemon_id=None) for row in q.yield_per(5): - row_key = dict(table=cls.__tablename__, row_id=row.id) + row_key = dict(table=cls.__tablename__, row_id=unicode(row.id)) - # Spelling index only indexes strings of letters, alas, so we - # reduce every name to this to make the index work. However, exact - # matches are not returned, so e.g. 'nidoran' would neither match - # exactly nor fuzzy-match. Solution: add the spelling-munged name - # as a regular index row too. name = row.name.lower() writer.add_document(name=name, **row_key) + speller_entries.append((name, 1)) - speller_entries.append(name) + # Pokemon also get other languages + for foreign_name in getattr(row, 'foreign_names', []): + moonspeak = foreign_name.name.lower() + if name == moonspeak: + # Don't add the English name again as a different language; + # no point and it makes spell results confusing + continue - for extra_key_func in extra_keys.get(cls, []): - extra_key = extra_key_func(row) - writer.add_document(name=extra_key, **row_key) + writer.add_document(name=moonspeak, + language=foreign_name.language.name, + **row_key) + speller_entries.append((moonspeak, 3)) + + # Add Roomaji too + if foreign_name.language.name == 'Japanese': + roomaji = romanize(foreign_name.name).lower() + writer.add_document(name=roomaji, language='Roomaji', + **row_key) + speller_entries.append((roomaji, 8)) - writer.commit() - # XXX GIHWEGREHKG - old__schema = whoosh.spelling.SpellChecker._schema - def new__schema(self): - schema = old__schema(self) - schema.add('dummy', whoosh.fields.TEXT) - return schema - whoosh.spelling.SpellChecker._schema = new__schema + 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 - speller = whoosh.spelling.SpellChecker(index.storage, indexname='spelling') - # WARNING: HERE BE DRAGONS - # whoosh.spelling refuses to index things that don't look like words. - # Unfortunately, this doesn't work so well for Pokémon (Mr. Mime, - # Porygon-Z, etc.), and attempts to work around it lead to further - # complications. - # The below is copied from SpellChecker.add_scored_words without the check - # for isalpha(). XXX get whoosh patched to make this unnecessary! - writer = speller.index(create=True).writer() - for word in speller_entries: - fields = {"word": word, "score": 1} - for size in xrange(speller.mingram, speller.maxgram + 1): - nga = whoosh.analysis.NgramAnalyzer(size) - gramlist = [t.text for t in nga(word)] - if len(gramlist) > 0: - fields["start%s" % size] = gramlist[0] - fields["end%s" % size] = gramlist[-1] - fields["gram%s" % size] = " ".join(gramlist) - writer.add_document(**fields) - writer.commit() - # end copy-pasta + speller = whoosh.spelling.SpellChecker(index.storage) + speller.add_scored_words(speller_entries) return index, speller -def lookup(name, session=None, exact_only=False): +class LanguageWeighting(whoosh.scoring.Weighting): + """A scoring class that forces otherwise-equal English results to come + before foreign results. + """ + + def score(self, searcher, fieldnum, text, docnum, weight, QTF=1): + doc = searcher.stored_fields(docnum) + if doc['language'] == None: + # English (well, "default"); leave it at 1 + return weight + elif doc['language'] == u'Roomaji': + # Give Roomaji a bit of a boost, as it's most likely to be searched + return weight * 0.95 + else: + # Everything else can drop down the totem pole + return weight * 0.9 + +rx_is_number = re.compile('^\d+$') + +LookupResult = namedtuple('LookupResult', + ['object', 'name', 'language', 'exact']) +def lookup(input, session=None, indices=None, exact_only=False): """Attempts to find some sort of object, given a database session and name. - Returns (objects, exact) where `objects` is a list of database objects, and - `exact` is True iff the given name matched the returned objects exactly. + Returns a list of named (object, name, language, exact) tuples. `object` + is a database object, `name` is the name under which the object was found, + `language` is the name of the language in which the name was found, and + `exact` is True iff this was an exact match. - This function ONLY does fuzzy matching if there are no exact matches. + This function currently ONLY does fuzzy matching if there are no exact + matches. Formes are not returned; "Shaymin" will return only grass Shaymin. - Currently recognizes: - - Pokémon names: "Eevee" + Recognizes: + - Names: "Eevee", "Surf", "Run Away", "Payapa Berry", etc. + - Foreign names: "Iibui", "Eivui" + - Fuzzy names in whatever language: "Evee", "Ibui" + - IDs: "pokemon 133", "move 192", "item 250" + - Dex numbers: "sinnoh 55", "133", "johto 180" - `name` + `input` Name of the thing to look for. `session` @@ -181,6 +187,10 @@ def lookup(name, session=None, exact_only=False): if this is not provided, a connection to the default database will be attempted. + `indices` + Tuple of index, speller as returned from `open_index()`. Defaults to + a call to `open_index()`. + `exact_only` If True, only exact matches are returned. If set to False (the default), and the provided `name` doesn't match anything exactly, @@ -190,27 +200,42 @@ def lookup(name, session=None, exact_only=False): if not session: session = connect() - index, speller = open_index() + if indices: + index, speller = indices + else: + index, speller = open_index() + name = unicode(input).lower() exact = True - # Look for exact name. A Term object does an exact match, so we don't have - # to worry about a query parser tripping on weird characters in the input + # If the input provided is a number, match it as an id. Otherwise, name. + # Term objects do an exact match, so we don't have to worry about a query + # parser tripping on weird characters in the input + if rx_is_number.match(name): + # Don't spell-check numbers! + exact_only = True + query = whoosh.query.Term(u'row_id', name) + else: + # Not an integer + query = whoosh.query.Term(u'name', name) + + ### Actual searching searcher = index.searcher() - query = whoosh.query.Term('name', name.lower()) + searcher.weighting = LanguageWeighting() # XXX kosher? docs say search() + # takes a weighting kw but it + # certainly does not results = searcher.search(query) - if not exact_only: - # Look for some fuzzy matches - if not results: - exact = False - results = [] + # Look for some fuzzy matches if necessary + if not exact_only and not results: + exact = False + results = [] - for suggestion in speller.suggest(name, 3): - query = whoosh.query.Term('name', suggestion) - results.extend(searcher.search(query)) + for suggestion in speller.suggest(name, 25): + query = whoosh.query.Term('name', suggestion) + results.extend(searcher.search(query)) - # Convert results to db objects + ### Convert results to db objects objects = [] seen = {} for result in results: @@ -222,6 +247,14 @@ def lookup(name, session=None, exact_only=False): cls = indexed_tables[result['table']] obj = session.query(cls).get(result['row_id']) - objects.append(obj) - - return objects, exact + objects.append(LookupResult(object=obj, + name=result['name'], + language=result['language'], + exact=exact)) + + # Only return up to 10 matches; beyond that, something is wrong. + # We strip out duplicate entries above, so it's remotely possible that we + # should have more than 10 here and lost a few. The speller returns 25 to + # give us some padding, and should avoid that problem. Not a big deal if + # we lose the 25th-most-likely match anyway. + return objects[:10]