As we can see, for Pittsburgh for example, 2 is the best number of clusters for the check-in matrix. We then select the k best features (from those 477 features) that affect the membership status of a restaurant in one of those previously defined clusters. In other words, we discover ...