Databases Reference
In-Depth Information
Leave the grid size unchanged.
Check the boxes to compute all three statistics.
Click OK.
The small iris dataset processes almost immediately. Larger datasets will take
more time. If processing time is an issue, you may not want to compute all statistics.
Computing the silhouette coefficient can be especially time consuming.
Drag the newly created SOM:IrisSize model up to a display and drop.
Select “SOM Viewer”.
The primary SOM viewer pane (Figure 7.7) represents the grid. Each cluster
in the grid is drawn as a sphere whose size corresponds to the number of
observations belonging to the cluster. To get an accurate assessment of the grid
and the cluster locations within the grid, it helps to rotate by dragging either up
and down or left and right. Note: The grid that you see when working with the
datasets in this chapter may mirror or reflect images of the published figures.
As the cluster analysis progresses, dominant clusters migrate to the corners of
the grid. The actual ending corner for a given cluster depends on the initial
random values assigned to the centroids.
The options panel (Figure 7.8) presents analysis statistics and supports some
grid formatting options. In this particular clustering, the overall mean squared
error (MSE) is 0.0423. This value only becomes meaningful when comparing
clusterings or when comparing the MSE of an individual against the overall
MSE. The clustering's coefficient of correlation is 0.516, a respectable value.
In the Transparency Options box, select the “MSE Base” radio button.
Figure 7.7
Iris SOM Grid
