The authors of the following paper predict the market attraction for restaurants based on user reviews for different locations to select an ideal location. To do so they have split the city into a set of candidate locations and scored each candidate location.
They explain the methodology as follows:
"To obtain the candidate locations L, we use a density-based spatial clustering method (OPTICS) to segment a city into cells, each of which is treated as a candidate location. As a result, in our experiment, New York city is segmented into 995 cells"
I have read on the scikit-learn implementation of OPTICS and as far as I understand it is similar to DBSCAN, while it can cluster latitude and longitudes I'm lost on how they have used it to split a city (bounding box) into a set of candidate locations, and given a lat/long point map it into a cell? How would they have split New York city into 995 cells?