Databases Reference
In-Depth Information
Figure 7.10
SOM Vote Clustering
To verify, click on each of the two clusters, then review in the synchronized
data table. (You may want to stretch the “RepName” column header to make
room for the full name.)
The clusters adjacent to the two dominant clusters contain representatives that
usually voted the party line yet diverged at times. This is where the advantage
of the 3-D SOM becomes evident. There are three clusters adjacent to the
“Republican” cluster. Each identifies a different pattern in the way they strayed.
The clusters in the cells distant from the two dominant clusters represent
the more independent voters. They did not vote the party lines, yet given the
distances between some of these independent clusters, their patterns of devia-
tion varied. Out of curiosity, you may want to click on each of these clusters to
see in the synchronized data table who these representatives are and the states
they come from.
To compare the voting records between clusters, close the data table and
open in its place a synchronized parallel plot of the “votes” dataset.
Use the “Ctrl-click”optionof the SOMto selectmultiple clusters to compare.
The default plot is ugly and difficult to interpret. Check “Show Means” to
get a more meaningful plot.
Without the three-dimensional grid, we would not have had that extra
dimension along which clusters could vary from the dominant corner clusters.
Extracting Subsets from a Clustering
The dataset OutdoorCustomerSales.csv contains a sampling of sales data for
online sales of Big Box Merchandising, a large warehouse retailer. The
Purchases column contains the total online purchases for the year of outdoor
