How are node attributes being exploited?

[muety]

The docs mention that the library supports to incorporate node attributes when computing the distance. Without diving deeply into the code, how is this implemented, that is, how are two feature vectors compared to yield a distance?