Figure 6From: Spectral embedding finds meaningful (relevant) structure in image and microarray dataWeighted values vs. Euclidean distance. For each plot, the black trend line is drawn based on ordering the values for each vector (weighted values and Euclidean distance values); this gives a general fitting of each curve. The portion of the rainbow spectrum (ROYGB) shading for the lines are drawn by the same criterion, however, each line is calculated based on a set of increasing epsilon smoothing values. For example, epsilon values are increased from the 1% quantile to the 50% quantile of the squared Euclidean distance distribution, where each set of 10% values are plotted with a separate color in the rainbow spectrum, staring at red and ending at blue. This method of line shading illustrates how the transformed Euclidean distances are adjusted across a dynamic range of epsilon values for each method. (a) Gaussian radial basis function kernel values vs. Euclidean distances. (b) Entries in the matrix of the weighted graph Laplacian, L from Ng et al., vs. Euclidean distance. (c) Entries in the matrix of the weighted graph Laplacian, K from Lafon vs. Euclidean distance.Back to article page