Make lookup ignore accent marks.

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

diff --git a/pokedex/lookup.py b/pokedex/lookup.py
index 16c4e5e..b31dce8 100644 (file)
--- a/pokedex/lookup.py
+++ b/pokedex/lookup.py
@@ -2,7 +2,10 @@
 from collections import namedtuple
 import os, os.path
 import pkg_resources
 from collections import namedtuple
 import os, os.path
 import pkg_resources

+import random

 import re
 import re

+import shutil
+import unicodedata

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


@@ -17,7 +20,10 @@ from pokedex.db import connect
 import pokedex.db.tables as tables
 from pokedex.roomaji import romanize
 
 import pokedex.db.tables as tables
 from pokedex.roomaji import romanize
 

-__all__ = ['open_index', 'lookup']
+__all__ = ['open_index', 'lookup', 'random_lookup']
+
+INTERMEDIATE_LOOKUP_RESULTS = 25
+MAX_LOOKUP_RESULTS = 10

 
 # 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


@@ -32,6 +38,28 @@ for cls in [
     ]:
     indexed_tables[cls.__tablename__] = cls
 
     ]:
     indexed_tables[cls.__tablename__] = cls
 

+def normalize(name):
+    """Strips irrelevant formatting junk from name input.
+
+    Specifically: everything is lowercased, and accents are removed.
+    """
+    # http://stackoverflow.com/questions/517923/what-is-the-best-way-to-remove-accents-in-a-python-unicode-string
+    # Makes sense to me.  Decompose by Unicode rules, then remove combining
+    # characters, then recombine.  I'm explicitly doing it this way instead of
+    # testing combining() because Korean characters apparently decompose!  But
+    # the results are considered letters, not combining characters, so testing
+    # for Mn works well, and combining them again makes them look right.
+    nkfd_form = unicodedata.normalize('NFKD', unicode(name))
+    name = u"".join(c for c in nkfd_form
+                    if unicodedata.category(c) != 'Mn')
+    name = unicodedata.normalize('NFC', name)
+
+    name = name.strip()
+    name = name.lower()
+
+    return name
+
+

 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.
 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.


@@ -53,7 +81,7 @@ def open_index(directory=None, session=None, recreate=False):
     # Defaults
     if not directory:
         directory = pkg_resources.resource_filename('pokedex',
     # Defaults
     if not directory:
         directory = pkg_resources.resource_filename('pokedex',

-                                                    'data/whoosh_index')
+                                                    'data/whoosh-index')

 
     if not session:
         session = connect()
 
     if not session:
         session = connect()


@@ -71,17 +99,27 @@ def open_index(directory=None, session=None, recreate=False):
             # Apparently not a real index.  Fall out of the if and create it
             pass
 
             # Apparently not a real index.  Fall out of the if 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)
 
 
     if not directory_exists:
         os.mkdir(directory)
 
 

-    # Create index
+    ### 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,
+        table=whoosh.fields.ID(stored=True),

         row_id=whoosh.fields.ID(stored=True),
         language=whoosh.fields.STORED,
         row_id=whoosh.fields.ID(stored=True),
         language=whoosh.fields.STORED,

+        iso3166=whoosh.fields.STORED,

         display_name=whoosh.fields.STORED,  # non-lowercased name
         display_name=whoosh.fields.STORED,  # non-lowercased name

+        forme_name=whoosh.fields.ID,

     )
 
     index = whoosh.index.create_in(directory, schema=schema, indexname='MAIN')
     )
 
     index = whoosh.index.create_in(directory, schema=schema, indexname='MAIN')


@@ -96,18 +134,27 @@ def open_index(directory=None, session=None, recreate=False):
     for cls in indexed_tables.values():
         q = session.query(cls)
 
     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=unicode(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')
+
+            def add(name, language, iso3166, score):
+                normalized_name = normalize(name)
+                writer.add_document(name=normalized_name, display_name=name,
+                                    language=language,
+                                    iso3166=iso3166,
+                                    **row_key)
+                speller_entries.append((normalized_name, score))
+
+            # If this is a form, mark it as such
+            if getattr(row, 'forme_base_pokemon_id', None):
+                row_key['forme_name'] = row.forme_name

 
             name = row.name
 
             name = row.name

-            writer.add_document(name=name.lower(),
-                                display_name=name,
-                                **row_key)
-            speller_entries.append((name.lower(), 1))
+            add(name, None, u'us', 1)

 
             # Pokemon also get other languages
             for foreign_name in getattr(row, 'foreign_names', []):
 
             # Pokemon also get other languages
             for foreign_name in getattr(row, 'foreign_names', []):


@@ -117,21 +164,14 @@ def open_index(directory=None, session=None, recreate=False):
                     # no point and it makes spell results confusing
                     continue
 
                     # 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(moonspeak, foreign_name.language.name,
+                               foreign_name.language.iso3166,
+                               3)

 
                 # Add Roomaji too
                 if foreign_name.language.name == 'Japanese':
                     roomaji = romanize(foreign_name.name)
 
                 # 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))
-
+                    add(roomaji, u'Roomaji', u'jp', 8)

 
     writer.commit()
 
 
     writer.commit()
 


@@ -164,30 +204,88 @@ class LanguageWeighting(whoosh.scoring.Weighting):
 rx_is_number = re.compile('^\d+$')
 
 LookupResult = namedtuple('LookupResult',
 rx_is_number = re.compile('^\d+$')
 
 LookupResult = namedtuple('LookupResult',

-                          ['object', 'name', 'language', 'exact'])
-def lookup(input, session=None, indices=None, exact_only=False):
+                          ['object', 'name', 'language', 'iso3166', 'exact'])
+
+def _parse_table_name(name):
+    """Takes a singular table name, table name, or table object and returns the
+    table name.
+
+    Returns None for a bogus name.
+    """
+    if hasattr(name, '__tablename__'):
+        return getattr(name, '__tablename__')
+    elif name in indexed_tables:
+        return name
+    elif name + 's' in indexed_tables:
+        return name + 's'
+    else:
+        # Bogus.  Be nice and return dummy
+        return None
+
+def _whoosh_records_to_results(records, session, exact=True):
+    """Converts a list of whoosh's indexed records to LookupResult tuples
+    containing database objects.
+    """
+    # XXX this 'exact' thing is getting kinda leaky.  would like a better way
+    # to handle it, since only lookup() cares about fuzzy results
+    seen = {}
+    results = []
+    for record in records:
+        # Skip dupes
+        seen_key = record['table'], record['row_id']
+        if seen_key in seen:
+            continue
+        seen[seen_key] = True
+
+        cls = indexed_tables[record['table']]
+        obj = session.query(cls).get(record['row_id'])
+
+        results.append(LookupResult(object=obj,
+                                    name=record['display_name'],
+                                    language=record['language'],
+                                    iso3166=record['iso3166'],
+                                    exact=exact))
+
+    return results
+
+
+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 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.
+    Returns a list of named (object, name, language, iso3166, exact) tuples.
+    `object` is a database object, `name` is the name under which the object
+    was found, `language` and `iso3166` are the name and country code of the
+    language in which the name was found, and `exact` is True iff this was an
+    exact match.

 
     This function currently 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 unless requested; "Shaymin" will return only grass
+    Shaymin.
+
+    Extraneous whitespace is removed with extreme prejudice.

 
     Recognizes:
     - Names: "Eevee", "Surf", "Run Away", "Payapa Berry", etc.
     - Foreign names: "Iibui", "Eivui"
     - Fuzzy names in whatever language: "Evee", "Ibui"
 
     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"
+    - 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.
 
 
     `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
     `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


@@ -211,56 +309,147 @@ def lookup(input, session=None, indices=None, exact_only=False):
     else:
         index, speller = open_index()
 
     else:
         index, speller = open_index()
 

-    name = unicode(input).lower()
+    name = normalize(input)

     exact = True
     exact = True

-
-    # 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):
+    form = None
+
+    # Remove any type prefix (pokemon:133) before constructing a query
+    if ':' in name:
+        prefix_chunk, name = name.split(':', 1)
+        name = name.strip()
+
+        if not valid_types:
+            # Only use types from the query string if none were explicitly
+            # provided
+            prefixes = prefix_chunk.split(',')
+            valid_types = [_.strip() for _ in prefixes]
+
+    # Random lookup
+    if name == 'random':
+        return random_lookup(indices=(index, speller),
+                             session=session,
+                             valid_types=valid_types)
+
+    # 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
+    if '*' in name or '?' in name:
+        exact_only = True
+        query = whoosh.query.Wildcard(u'name', name)
+    elif 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
         # 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)
+        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(' ', 1)
+            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:
+        table_name = _parse_table_name(valid_type)
+        if table_name:
+            # Quietly ignore bogus valid_types; more likely to DTRT
+            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.weighting = LanguageWeighting()  # XXX kosher?  docs say search()
                                               # takes a weighting kw but it
                                               # certainly does not
 
     ### Actual searching
     searcher = index.searcher()
     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, limit=INTERMEDIATE_LOOKUP_RESULTS)

 
     # Look for some fuzzy matches if necessary
     if not exact_only and 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, 25):
+        for suggestion in speller.suggest(name, INTERMEDIATE_LOOKUP_RESULTS):

             query = whoosh.query.Term('name', suggestion)
             results.extend(searcher.search(query))
 
     ### Convert results to db objects
             query = whoosh.query.Term('name', suggestion)
             results.extend(searcher.search(query))
 
     ### Convert results to db objects

-    objects = []
-    seen = {}
-    for result in results:
-        # Skip dupe results
-        seen_key = result['table'], result['row_id']
-        if seen_key in seen:
-            continue
-        seen[seen_key] = True
-
-        cls = indexed_tables[result['table']]
-        obj = session.query(cls).get(result['row_id'])
-        objects.append(LookupResult(object=obj,
-                                    name=result['display_name'],
-                                    language=result['language'],
-                                    exact=exact))
+    objects = _whoosh_records_to_results(results, session, 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.
 
     # 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]
+    return objects[:MAX_LOOKUP_RESULTS]
+
+
+def random_lookup(valid_types=[], session=None, indices=None):
+    """Takes similar arguments as `lookup()`, but returns a random lookup
+    result from one of the provided `valid_types`.
+    """
+
+    tables = []
+    for valid_type in valid_types:
+        table_name = _parse_table_name(valid_type)
+        if table_name:
+            tables.append(indexed_tables[table_name])
+
+    if not tables:
+        # n.b.: It's possible we got a list of valid_types and none of them
+        # were valid, but this function is guaranteed to return *something*, so
+        # it politely selects from the entire index isntead
+        tables = indexed_tables.values()
+
+    # Rather than create an array of many hundred items and pick randomly from
+    # it, just pick a number up to the total number of potential items, then
+    # pick randomly from that, and partition the whole range into chunks.
+    # This also avoids the slight problem that the index contains more rows
+    # (for languages) for some items than others.
+    # XXX ought to cache this (in the index?) if possible
+    total = 0
+    partitions = []
+    for table in tables:
+        count = session.query(table).count()
+        total += count
+        partitions.append((table, count))
+
+    n = random.randint(1, total)
+    while n > partitions[0][1]:
+        n -= partitions[0][1]
+        partitions.pop(0)
+
+    return lookup(unicode(n), valid_types=[ partitions[0][0] ],
+                  indices=indices, session=session)
+
+def prefix_lookup(prefix, session=None, indices=None):
+    """Returns terms starting with the given exact prefix.
+
+    No special magic is currently done with the name; type prefixes are not
+    recognized.
+
+    `session` and `indices` are treated as with `lookup()`.
+    """
+
+    if not session:
+        session = connect()
+
+    if indices:
+        index, speller = indices
+    else:
+        index, speller = open_index()
+
+    query = whoosh.query.Prefix(u'name', normalize(prefix))
+
+    searcher = index.searcher()
+    searcher.weighting = LanguageWeighting()
+    results = searcher.search(query)  # XXX , limit=MAX_LOOKUP_RESULTS)
+
+    return _whoosh_records_to_results(results, session)