Java Reference
In-Depth Information
In addition, not all vendors implement or expose the same level of
DME capabilities. JDM is designed to be flexible to adapt for any
vendor that supports at least one data mining function. Moreover,
JDM enables applications to discover DME capabilities at runtime.
This feature is particularly useful for the development of vendor-
neutral data mining tools or solutions.
To reduce the complexity of the implementation and integration
of data mining solutions, DMEs are providing advanced automation
for data preparation, attribute selection, algorithm selection, and
automated tuning of algorithm settings. This simplifies data mining
for novice users, allowing them to quickly reach a solution. The
expert group designed JDM to be flexible for defining both auto-
mated data mining processes and highly customized solutions
involving problem-specific settings by expert users.
This chapter discusses some of the design decisions made for
JDM. Readers will also become familiar with the JDM object model,
packages, and objects. This chapter uses Unified Modeling Lan-
guage [OMG_UML 2006] notation for class diagrams and package
diagrams.
8.1
Object Modeling of Data Mining Concepts
This section explores the JDM object model for representing the data
mining concepts discussed in Chapter 7: data specifications, the set-
tings used for data mining techniques and algorithms, function level
model contents, algorithm level model details, test metrics for
supervised models, and model apply. The JDM object model enables
developing data mining solutions using Java and Web services inter-
faces.
Figure 8-1 depicts the class diagram of mining objects in JDM—a
mining object in JDM is the prime object that can be saved with a
name to the mining object repository (MOR) via the DME. Applica-
tions can then retrieve mining objects by name from the MOR across
application executions. JDM divides mining objects into data specifi-
cation, settings, model, test metrics, and task objects. The gray boxes
in Figure 8-1 highlight the mining object categories Data Specification
Objects and Settings Objects . The MiningObject interface encapsulates
the common characteristics of these objects, such as name, descrip-
tion, object identifier, and the mining object type. We now explore
each mining object category in detail.
Search WWH ::




Custom Search