Search code examples
algorithmcluster-analysisdbscanoptics-algorithm

Is radius epsilon inclusive in DBSCAN/OPTICS algorithms?


The original paper and other resources (wikipedia) always define a core-object depending on a radius ε (there must be more than MinPts neighbors) using expressions like within or up to. This leaves a room for interpretation whether this radius is inclusive or not: is an object q a neighboor of p if the distance(p, q) is exaclty ε?

It is obviously very important since all the definitions are based on that...


Solution

  • The definition in the original dbscan paper for the neighborhood is dist(p, q) <= eps. However, most implementation will use whatever range query the database index supports. In most cases, it will not make a difference and OmG is right that by adding the smallest representable number to eps will effectively change < to <=.