Added lookup tests. #15

[zzz-pokedex.git] / pokedex / lookup.py

diff --git a/pokedex/lookup.py b/pokedex/lookup.py
index 8c92edb..51bbc4e 100644 (file)
--- a/pokedex/lookup.py
+++ b/pokedex/lookup.py
@@ -1,12 +1,23 @@
 # encoding: utf8
 # encoding: utf8

+from collections import namedtuple
+import os, os.path
+import pkg_resources

 import re
 
 from sqlalchemy.sql import func
 import whoosh
 import re
 
 from sqlalchemy.sql import func
 import whoosh

+import whoosh.filedb.filestore
+import whoosh.filedb.fileindex
+import whoosh.index

 from whoosh.qparser import QueryParser
 from whoosh.qparser import QueryParser

+import whoosh.scoring

 import whoosh.spelling
 
 import whoosh.spelling
 

+from pokedex.db import connect

 import pokedex.db.tables as tables
 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
 
 # Dictionary of table name => table class.
 # Need the table name so we can get the class from the table name after we


@@ -21,64 +32,111 @@ for cls in [
     ]:
     indexed_tables[cls.__tablename__] = cls
 
     ]:
     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),
-    ],
-}
-
-index_bits = {}
-def get_index(session):
-    """Returns (index, speller).
-
-    Creates an index if one does not exist.
+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.
+
+    `directory`
+        Directory containing the index.  Defaults to a location within the
+        `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.

     """
 
     """
 

-    if index_bits:
-        return index_bits['index'], index_bits['speller']
+    # Defaults
+    if not directory:
+        directory = pkg_resources.resource_filename('pokedex',
+                                                    'data/whoosh_index')
+
+    if not session:
+        session = connect()

 
 

-    store = whoosh.store.RamStorage()
+    # 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!
+        try:
+            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
+            pass
+
+    if not directory_exists:
+        os.mkdir(directory)
+
+
+    ### Create index

     schema = whoosh.fields.Schema(
         name=whoosh.fields.ID(stored=True),
     schema = whoosh.fields.Schema(
         name=whoosh.fields.ID(stored=True),

-        table=whoosh.fields.STORED,
-        row_id=whoosh.fields.STORED,
-        language_id=whoosh.fields.STORED,
+        table=whoosh.fields.ID(stored=True),
+        row_id=whoosh.fields.ID(stored=True),
+        language=whoosh.fields.STORED,
+        display_name=whoosh.fields.STORED,  # non-lowercased name
+        forme_name=whoosh.fields.ID,

     )
 
     )
 

-    # Construct a straight lookup index
-    index = whoosh.index.Index(store, schema=schema, create=True)
+    index = whoosh.index.create_in(directory, schema=schema, indexname='MAIN')

     writer = index.writer()
 
     # Index every name in all our tables of interest
     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)
 
     speller_entries = []
     for cls in indexed_tables.values():
         q = session.query(cls)
 

-        # Only index base Pokémon formes
-        if hasattr(cls, 'forme_base_pokemon_id'):
-            q = q.filter_by(forme_base_pokemon_id=None)
-

         for row in q.yield_per(5):
         for row in q.yield_per(5):

-            row_key = dict(table=cls.__tablename__, row_id=row.id)
+            # XXX need to give forme_name a dummy value because I can't search
+            # for explicitly empty fields.  boo.
+            row_key = dict(table=unicode(cls.__tablename__),
+                           row_id=unicode(row.id),
+                           forme_name=u'XXX')

 
 

-            # 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)
+            # If this is a form, mark it as such
+            if getattr(row, 'forme_base_pokemon_id', None):
+                row_key['forme_name'] = row.forme_name

 
 

-            speller_entries.append(name)
+            name = row.name
+            writer.add_document(name=name.lower(),
+                                display_name=name,
+                                **row_key)
+            speller_entries.append((name.lower(), 1))
+
+            # Pokemon also get other languages
+            for foreign_name in getattr(row, 'foreign_names', []):
+                moonspeak = foreign_name.name
+                if name == moonspeak:
+                    # Don't add the English name again as a different language;
+                    # no point and it makes spell results confusing
+                    continue
+
+                writer.add_document(name=moonspeak.lower(),
+                                    language=foreign_name.language.name,
+                                    display_name=moonspeak,
+                                    **row_key)
+                speller_entries.append((moonspeak.lower(), 3))
+
+                # Add Roomaji too
+                if foreign_name.language.name == 'Japanese':
+                    roomaji = romanize(foreign_name.name)
+                    writer.add_document(name=roomaji.lower(),
+                                        language='Roomaji',
+                                        display_name=roomaji,
+                                        **row_key)
+                    speller_entries.append((roomaji.lower(), 8))

 
 

-            for extra_key_func in extra_keys.get(cls, []):
-                extra_key = extra_key_func(row)
-                writer.add_document(name=extra_key, **row_key)

 
     writer.commit()
 
 
     writer.commit()
 


@@ -86,67 +144,156 @@ def get_index(session):
     # at once, as every call to add_* does a commit(), and those seem to be
     # expensive
     speller = whoosh.spelling.SpellChecker(index.storage)
     # at once, as every call to add_* does a commit(), and those seem to be
     # expensive
     speller = whoosh.spelling.SpellChecker(index.storage)

-    # 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 = whoosh.writing.IndexWriter(speller.index())
-    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.add_scored_words(speller_entries)
+
+    return index, speller
+
+
+class LanguageWeighting(whoosh.scoring.Weighting):
+    """A scoring class that forces otherwise-equal English results to come
+    before foreign results.
+    """

 
 

-    index_bits['index'] = index
-    index_bits['speller'] = speller
-    index_bits['store'] = store
-    return index_bits['index'], index_bits['speller']
+    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

 
 

-def lookup(session, name, exact_only=False):
+rx_is_number = re.compile('^\d+$')
+
+LookupResult = namedtuple('LookupResult',
+                          ['object', 'name', 'language', 'exact'])
+def lookup(input, valid_types=[], session=None, indices=None, exact_only=False):

     """Attempts to find some sort of object, given a database session and name.
 
     """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.
 
 
     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: "133", "192", "250"
+    Also:
+    - Type restrictions.  "type:psychic" will only return the type.  This is
+      how to make ID lookup useful.  Multiple type specs can be entered with
+      commas, as "move,item:1".  If `valid_types` are provided, any type prefix
+      will be ignored.
+    - Alternate formes can be specified merely like "wash rotom".
+
+    `input`
+        Name of the thing to look for.
+
+    `valid_types`
+        A list of table objects or names, e.g., `['pokemon', 'moves']`.  If
+        this is provided, only results in one of the given tables will be
+        returned.
+
+    `session`
+        A database session to use for retrieving objects.  As with get_index,
+        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,
+        spelling correction will be attempted.

     """
 
     """
 

+    if not session:
+        session = connect()
+
+    if indices:
+        index, speller = indices
+    else:
+        index, speller = open_index()
+
+    name = unicode(input).lower()

     exact = True
     exact = True

+    form = None

 
 

-    index, speller = get_index(session)
+    # Remove any type prefix (pokemon:133) before constructing a query
+    if ':' in name:
+        prefix_chunk, name = name.split(':', 2)
+        prefixes = prefix_chunk.split(',')
+        if not valid_types:
+            # Only use types from the query string if none were explicitly
+            # provided
+            valid_types = prefixes

 
 

-    # 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) \
+              & whoosh.query.Term(u'forme_name', u'XXX')
+
+        # If there's a space in the input, this might be a form
+        if ' ' in name:
+            form, formless_name = name.split(' ', 2)
+            form_query = whoosh.query.Term(u'name', formless_name) \
+                       & whoosh.query.Term(u'forme_name', form)
+            query = query | form_query
+
+    ### Filter by type of object
+    type_terms = []
+    for valid_type in valid_types:
+        if hasattr(valid_type, '__tablename__'):
+            table_name = getattr(valid_type, '__tablename__')
+        elif valid_type in indexed_tables:
+            table_name = valid_type
+        elif valid_type + 's' in indexed_tables:
+            table_name = valid_type + 's'
+        else:
+            # Bogus.  Be nice and ignore it
+            continue
+
+        type_terms.append(whoosh.query.Term(u'table', table_name))
+
+    if type_terms:
+        query = query & whoosh.query.Or(type_terms)
+
+
+    ### Actual searching

     searcher = index.searcher()
     searcher = index.searcher()

-    query = whoosh.query.Term('name', name)
+    searcher.weighting = LanguageWeighting()  # XXX kosher?  docs say search()
+                                              # takes a weighting kw but it
+                                              # certainly does not

     results = searcher.search(query)
 
     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:
     objects = []
     seen = {}
     for result in results:


@@ -158,6 +305,15 @@ def lookup(session, name, exact_only=False):
 
         cls = indexed_tables[result['table']]
         obj = session.query(cls).get(result['row_id'])
 
         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['display_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]