Graphics Reference
In-Depth Information
Figure
.
.
Variable and unit representations ater the deletion of the most important units
VITAMIN-S was expressly designed toperform exploratory analysis also on large
andverylargedatasets,soitcanexploithierarchicalandnonhierarchicalclusteranal-
ysis algorithms. he current example, consistent with the structure of the contribu-
tion, is realized using an agglomerative hierarchical approach: Ward's classification
algorithm. As described in Sect.
.
, cluster analysis assumes as input data the unit
coordinates on the factors (the principal components). Finally, the exploratory pro-
cessendswithadendrogramdisplay(let-handsideinFig.
.
),whereastheanalysis
processgoesontoobtain clusterdescriptions. Inthis phasethereareseveral different
data and analysis parameters the user can set. Looking at the dendrogram, we notice
the yellow vertical line. Moving the line from right to let, the user can set the classi-
fication tree cutting point and the number of partition classes as a consequence. At
the same time, on another display there is the cluster display on the factorial plan.
he user can act on both displays. In particular, when a cluster on the factorial rep-
resentation is deleted, the set of active units changes and the whole analysis process
works in the background to recalculate all the results of the entire analysis. Results
are immediately displayed in the updated views.
Finally, when the user is satisfied with the obtained configurations, the user can
click on a command button to save all results, graphical and analytical.
Conclusion and Perspectives
4.9
Recent visualization techniques are based on the highly intensive exploitation of PC
graphics cards capabilities. We should expect that these capabilities will further im-
prove in coming years. However, the risk is that these representations, which are
very attractive from an aesthetic point of view, will not be readable using a statis-
tical approach. Statistical reading of graphical representations must necessarily be
