Let’s look at an example of buildingan Artist Preference Modelthat predicts how likely a user is a fan of a given artist. If done in batch, this would require an inefficient number of predictions, roughly about (300M users x 10M artists) pairs. If instead be done online at request tim...