How to sum count subqueries with SQLAlchemy?

Question

I have the following models in my DB (Flask-SQLALchemy, declarative approach, simplified):

class Player(db.Model):
    id = db.Column(db.Integer, primary_key = True)
    ...

class Game(db.Model):
    id = db.Column(db.Integer, primary_key = True)
    creator_id = db.Column(db.Integer, db.ForeignKey('player.id'))
    creator = db.relationship(Player, foreign_keys='Game.creator_id')
    opponent_id = db.Column(db.Integer, db.ForeignKey('player.id'))
    opponent = db.relationship(Player, foreign_keys='Game.opponent_id')
    winner = db.Column(db.Enum('creator', 'opponent'))

Each game may be either won, lost or ended in draw. I need to get players sorting them by "win rate" - i.e.:

if player created a game and that game's winner is creator, it is considered win;
if the player was invited to game as opponent and game's winner is opponent, it is also considered win;
other games where this player participated are considered lost games.

So my algorithm is as follows:

@hybrid_property
def winrate(self):
    games = Game.query.filter(or_(
        Game.creator_id == self.id,
        Game.opponent_id == self.id,
    ))
    count = 0
    wins = 0
    for game in games:
        count += 1
        if game.creator_id == self.id and game.winner == 'creator':
            wins += 1
        elif game.opponent_id == self.id and game.winner == 'opponent':
            wins += 1
    if count == 0:
        return 0
    return wins / count

This approach works when I want to determine win rate for particular player; but it fails when I want to sort players by win rate. I tried to rewrite it in SQL and got something like this:

SELECT * FROM player
ORDER BY ((SELECT count(g1.id) FROM game g1
    WHERE g1.creator_id = player.id AND g1.winner = 'creator'
) + (SELECT count(g2.id) FROM game g2
    WHERE g2.opponent_id = player.id AND g2.winner = 'opponent'
)) / (SELECT count(g3.id) FROM game g3
    WHERE player.id IN (g3.creator_id, g3.opponent_id)
)

This doesn't handle players without games but should work in general. Players without games can be probably handled with MySQL CASE statement.

But the problem is that I cannot figure how do I encode this SQL using SQLAlchemy. Here is a (simplified) code I try to use:

@winrate.expression
def winrate(cls):
    cnt = Game.query.filter(
        cls.id.in_(Game.creator_id, Game.opponent_id)
    ).with_entities(func.count(Game.id))
    won = Game.query.filter(
        or_(
            and_(
                Game.creator_id == cls.id,
                Game.winner == 'creator',
            ),
            and_(
                Game.opponent_id == cls.id,
                Game.winner == 'opponent',
            ),
        )
    )
    return case([
        (count == 0, 0),
    ], else_ = (
        won / count
    ))

This code fails when it comes to won / count line telling me that Query cannot be divided by Query. I tried using subqueries but without any success.

How should I implement it? Or maybe I should use some kind of joins/whatever? (DB scheme cannot be changed.)

without actually looking or thinking too hard about this: won.scalar() / count.scalar()? Query.scalar() produces a scalar subquery, which is a clause element — Eevee
– Eevee, Commented Sep 6, 2015 at 3:08

van · Accepted Answer · 2015-09-07 09:34:24Z

1

Try working with core expressions instead of orm queries:

class Player(..):
    # ...
    @winrate.expression
    def _winrate(cls):
        cnt = (
            select([db.func.count(Game.id)])
            .where(
                db.or_(
                    Game.creator_id == cls.id,
                    Game.opponent_id == cls.id,
                ))
            .label("cnt")
        )
        won = (
            select([db.func.count(Game.id)])
            .where(
                db.or_(
                    db.and_(Game.creator_id == cls.id,
                            Game.winner == 'creator'),
                    db.and_(Game.opponent_id == cls.id,
                            Game.winner == 'opponent'),
                ))
            .label("cnt")
        )

        return db.case(
            [(cnt == 0, 0)],
            else_ = db.cast(won, db.Numeric) / cnt
        )
# ...
q = session.query(Player).order_by(Player.winrate.desc())

answered Sep 7, 2015 at 9:34

van

77.6k13 gold badges174 silver badges179 bronze badges

Sign up to request clarification or add additional context in comments.

1 Comment

MarSoft Over a year ago

Thanks very much! That worked. My only change was to use db.select instead of just select, but should not be critical.

Collectives™ on Stack Overflow

How to sum count subqueries with SQLAlchemy?

1 Answer 1

1 Comment

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

1 Comment

Your Answer

Sign up or log in

Post as a guest

Related