Graphics Reference
In-Depth Information
to see how everything is related at the big-picture level. It is also convenient
for summarizing important things about each subject in the same context.
Here, the distribution of values for each species is drawn in alternating
homogenous and heterogeneous rings, summarizing the defining
characteristics of each species. Clear radial striations in the Petal Length
and Petal Width nodes indicate that they are good features for classifying
flowers.
Because graphs can summarize relationships so effectively and so
efficiently, they can more easily scale to allow you to show more
information. For example, the number of features could easily be tripled in
this case, and the big picture would still be evident (and more interesting)
in
Figure 2-3
,
easily outdistancing the effectiveness of the matrix charts for
doing the same. Graphs are truly unparalleled in their capability to express
interconnected relationships.
Note
Chapter 9, “Relationships,” provides further examples of visualization
and analysis of relationships using graphs.
Hierarchies
Graphs are also a great choice for gaining insights from hierarchical data.
Hierarchical graphs are typically referred to as
trees
. Trees have a root
parent node with links branching to a second order of nodes, which may in
turn branch again, eventually reaching the leaf nodes that have no children.
Each node descendant of the root has a single parent.
Trees have many business applications.
Figure 2-4
reframes the Iris flower
classification information as a
decision tree
. A decision tree shows
sequences of decisions that lead to particular conclusions. Each node in
the tree is a decision, and each link represents a path to follow based on
particular criteria.
The Iris decision tree starts with the greatest distinguishing characteristic
of each species, which is petal length. All 50 of the Setosa samples can be
correctly identified by their characteristically short petals. If the petals are
longer, petal width can be measured and an estimate made as to whether
