Vendor Implementations - Java Data Mining: Strategy, Standard, and Practice - page 427

16.3 Guidelines for New Implementers

Guidelines for implementers 1 in developing a JDM implementation
are described in Section 6 of the JDM specification. This section
focuses on standards conformance, both in terms of a minimum JDM
implementation and vendor extensions, and using the TCK.

16.3.1 Standards Conformance

Conformance to the JDM standard is easier than for many other
standards due to JDM's á là carte capabilities mechanism. This mech-
anism allows implementing mainly the features a given product
already supports. The expert group designed JDM with this in mind
to lower the barrier to entry for new implementations. For example, a
data mining vendor with a single decision tree algorithm may decide
to offer only this algorithm as part of the classification mining
function.
Of course, a JDM API implementation is written in Java. How-
ever, the DME itself need not be written in Java. As noted earlier,
OJDM is a Java wrapper on top of data mining functions accessed
through SQL, and KJDM is a Java wrapper on top of a C++ or
CORBA library. The effort required to implement a Java wrapper on
top of existing engines can be minimal: Implementers are encour-
aged to start from their existing product lines and expose them
through JDM to minimize development effort.
The notion of capabilities allows implementers to adopt a phased
approach to JDM implementation, perhaps offering a first version for
a reduced feature set and then expanding the feature set. Implement-
ers can shape their JDM implementation using the capabilities and
the extension mechanism.
All JDM implementations share some core JDM functionality. The
next section describes a “minimal” JDM implementation.

1 The term “implementers” includes both vendors of data mining tools and

technology, and non-vendors. Non-vendors may include university students,
researchers, or companies with internal data mining capabilities.