Fig. 1

Fit of the distance transform functions to true coreness. The blue and red curves track on the vertical axis the average true coreness of training examples, whose nearest neighbor is respectively from a core or non-core class, at their corresponding nearest-neighbor distance on the horizontal axis. Average coreness is computed as described earlier with count \(\ell = 100\). The top and bottom green curves respectively show the fitted logistic functions \(f_\text {core}\) and \(f_\text {non}\) that are used by the coreness regressor to transform nearest-neighbor distance into predicted coreness