Practical Predictive Analytics

Graphically display cluster assignment

Bivariate cluster plots are often useful for seeing how the cluster assignments correlate with an x-y plot of two variables. Each cluster is plotted in a different color.

In Databricks, you can do this easily:

First, run the display command on some of the fitted data. In the code below, I have first extracted a 1,000-row sample. You want the sample to be small enough so that the points on the plot are not too dense.

             tmp <- head(sample(fitted, F, .01),1000)             display(tmp)              #show cluster assignment by 2 variable matrix 

Next, switch to the plot dialog box, Open the Customize Plot dialog. and perform the following graph setup:

1. Change the graph type to scatter plot.
2. Drag the prediction to the keys area. ...