whether or not we then out-of-sample embed the remaining vertices is not so important to me,
because i won't trust the oos so much anyway, because those vertices will have such a low degree.
also, a nice feature about this, with regard to the shinyapp, is that one could imagine a slider changing K
i'm thinking, in particular, a k-core decomposition prior to embedding, a la this paper: http://arxiv.org/pdf/1403.2226.pdf
whether or not we then out-of-sample embed the remaining vertices is not so important to me, because i won't trust the oos so much anyway, because those vertices will have such a low degree.
also, a nice feature about this, with regard to the shinyapp, is that one could imagine a slider changing K