Graphics Reference
In-Depth Information
Non-Cartesian Coordinates
here are many examples of useful plots and diagrams that require non-cartesian
coordinates, so it is desirable for statistical graphics sotware to support or at least
allow the construction of a variety of coordinate systems. For example, a number of
data sources suit polar coordinate displays, such as wind diagrams; when plotting
a single categorical variable with exactly three levels, ternary plots can be effective;
hierarchical data are naturally displayed as trees or graphs (with nodes and edges).
The User Interface
3.2.4
he user interface for providing parameters to control graphical output can be ade-
quatelyprovidedbyeitheracommandlineorGUI.Inacommand-lineenvironment,
function calls can be made with an argument provided for each control parameter;
GUIs tend to provide dialog boxes full of various options.
One issue that arises with statistical graphics is the 'explosion' of parameters for
higher-level graphical elements. Consider a matrix of scatterplots: the matrix con-
tains many plots; each plot contains several axes; each axis consists of multiple lines
and pieces of text. How can you provide parameters to control each piece of text in
every axis on every plot? hat is a lot of parameters. he problem essentially is one
of being able to uniquely specify a particular component of an overall plot.
A mouse input device provides a very good way of specifying elements in an im-
age.Itisverynaturaltopointattheelementyouwant.However,thereareissueswhen
selecting components of a plot because there is oten ambiguity due to the hierarchi-
cal structureinherent inaplot.Ifyouclickon apiece oftextonan axis tickmark,itis
not clear whether you want toselect just the text,or the entire axis, oreven the entire
plot. headvantage of using a command line toselect objects is that, although it may
be less convenient, you can typically be more expressive, or more precise. For exam-
ple, in the grid graphics system in R, the text for a particular axis might be expressed
as the following 'path':
"plot1::xaxis::label"
.
Another problem with the GUI approach is that it is hard to capture a particular
editing operation. For example, if the same editing operation is required on another
plot, the same series of actions must be repeated by the user. In a command-line
environment, operations can be captured and repeated easily.
Extensibility
3.3
he ability to produce complete plots, control all aspects of their appearance and
add additional output represents a minimum standard for what statistical graphics
sotware should provide. A more advanced feature is the ability to extend the system
to add new capabilities, such as new types of plots.
Insomerespects,creatinganewsortofplotisjustanextremeversionofcus-
tomization, but there are two distinguishing features: you are starting from a blank
slateratherthan buildingonanexisting plotasastarting point(i.e.itisnotjustanno-
tation) and, more importantly, extensibility means that the new plot that you create
