It looks like parallelizing over factors is a common thing to do, and we can attempt that and compare with our code. If we just let each thread do one factor per rating, that's too little work per thread. If we let each thread loop over all the users, that's too sequential. The optimal thing is probably somewhere in between.
It looks like parallelizing over factors is a common thing to do, and we can attempt that and compare with our code. If we just let each thread do one factor per rating, that's too little work per thread. If we let each thread loop over all the users, that's too sequential. The optimal thing is probably somewhere in between.