Figure 3.

Capability of the visualizations to preserve the similarities (the neighborhoods of size k) of the original data space. Sammon: Sammon's mapping, NMDS: non-metric multidimensional scaling, SOM: self-organizing map, HC: hierarchical clustering, with the ultrametric distance measure and with the linear distance measure. RP: Random linear projection is the approximate worst possible practical result (the small standard deviation over different projections, about 0.01, is not shown). The theoretical worst case, estimated with random neighborhoods, is approximately M₂ = 0.5. a) Yeast data. b) Mouse data.