Graphics Reference
In-Depth Information
Similar to how diagrams can be drawn informally by hand, diagrammatic
graphs can be generated formally by computers. Formalisms vary by
approach, but essentially, in any formal graph structure, subjects and
objects are represented by nodes, and relations are expressed by links.
When the goal is to understand the elements of a world and their
relationships, as well as
how
they are related, graphs are an invaluable
technique.
The representation of a relationship in a graph can be reduced to a line,
sometimes with a particular weight to indicate strength or volume. But in
reality, the underlying relationship often has more nuanced or expansive
characteristics than can be shown with a simple line. If the world being
displayed is reasonably small, visually expressive links, along with their
nodes, can help to more fully explain the nature of relationships.
One type of relationship that is fundamental to data science in virtually
any business is
correlation
. Correlations provide an indication of when
and how aspects of a world are related, which can inform decisions in
pursuit of business objectives. Understanding what conditions are most
favorable to a particular outcome provides the basis of a strategy for action,
influencing the probability of a profitable outcome by manipulating those
factors that are within control. Depending on the industry, that strategy
might take the form of targeted advertising, adjusting premiums based on a
risk assessment, or other actions.
Figure 2-1
reveals feature relationships in a modern take on a classic data
science study known as the Iris flower data set published by Sir Isaac Fisher
in 1936. A technique known as a
scatterplot matrix
is used to plot 50
samples of each of three species of Iris, for each pairwise combination of
four features. The features plotted in each scatterplot are found by following
the row and column to the feature labels. The data here represents flower
classificationsandtheirfeatures,butitcouldjustaswellrepresentcustomer
classifications and their purchasing or risk characteristics.
