Why does spark-ml ALS model returns NaN and negative numbers predictions?

Actually I'm trying to use ALS from spark-ml with implicit ratings.

I noticed that some predictions given by my trained model are negative or NaN, why is it?