Fig. 3
From: MetaRF: attention-based random forest for reaction yield prediction with a few trails
Example of nonlinear dimension reduction on the “Swiss roll” dataset. A Euclidean distance (dashed line) in the high-dimensional input space may not reflect the true low-dimensional geometry of the manifold (Solid line). B Using KS algorithm to select the most representative samples on the high-dimensional input space, samples in the central area (dashed square) will not be selected. C Using KS algorithm after the nonlinear dimension reduction, samples in the the central area (dashed square) will be selected