s-CorrPlot : Encoding and Exploring Correlation

Figure 1: Visualizations of correlation for a dataset containing 22,000 variables. The left two images show the correlation coefficients using a heatmap, clustered with (a) average linkage, and (b) complete linkage. The visible patterns in the heatmap are highly dependent on the clustering algorithm. In (c), our novel s-CorrPlot spatially encodes correlation coefficients, highlighting very different multidimensional structures in the data. The s-CorrPlot effectively encodes correlation through gridlines as in (d) for a specific variable of interest, p.