Model building and model apply involve the specification of various
settings. JDM allows vendor-specific defaults for settings that users
can use or override. Default values are assigned as part of settings
object creation and applications can retrieve the vendor-specified
default values using various get methods. This section discusses the
settings objects available to control the build and apply process.
To build models, users specify settings at the mining function level,
and optionally at the algorithm level. A mining function can be sup-
ported by multiple algorithms, similarly, the same algorithm can
support multiple mining functions. For example, a classification
function can use any of the algorithms such as naïve bayes, decision
tree, support vector machine, or neural networks. However, many of
these algorithms can also support the regression function. From a
design perspective, separating function settings from algorithm set-
tings reduces redundant specification of algorithm settings per func-
tion, but also provides more flexibility to allow users to specify
function-level settings only.
Figure 8-4 depicts the build settings—related objects and their
relationships. Here, the BuildSettings object is used for function-level
settings and the AlgorithmSettings object is used for algorithm-level
Build settings objects and their relationships.