When working with many time series a I'd like to be able to combine nested and global models to have a top performing ensemble (per id).