predicted_points ArrayImpl incompatible with postgresql

[Tdarnell]

The calc_predicted_points_for_player returned predicted_points is of type either None, float or jaxlib ArrayImpl.

The jaxlib ArrayImpl is incompatible with postgresql SQLAlchemy, and results in an error when trying to commit the predictions.

One option to fix this is:

for p in predictions:
        # check if db uri contains postgresql
        if "postgresql" in dbsession.bind.url.drivername:
            # check if the predicted_points is a float or jaxlib ArrayImpl
            if hasattr(p.predicted_points, "shape"):
                p.predicted_points = p.predicted_points.tolist()
        dbsession.add(p)

[jack89roberts]

If this is the case it's a bug that should be fixed elsewhere in calc_predicted_points_for_player, as PlayerPrediction.predicted_points should always be a non-null float value per the database schema (airsenal.framework.schema.PlayerPrediction), i.e. calc_predicted_points_for_player should always return a list of valid PlayerPrediction instances compatible with the DB schema.

From a very quick look I struggled to identify when/where the points could end up being a jax array, though. It might be another unintended consequence of us updating package versions recently.