7. <xsd:element name="validationDataMap" type="AttributeNameMap" minOccurs="0"
8. <xsd:attribute name="modelName" type="xsd:string" use="required"/>
9. <xsd:attribute name="inputModelName" type="xsd:string" use="optional"/>
10. <xsd:attribute name="applicationName" type="xsd:string" use="optional"/>
In the API, task input objects are always referenced by name, so
input objects must be saved and made available for task execution.
Since the XML Schema gives the choice of specifying task input
object contents, the task element can encapsulate all required inputs
to execute the task with no object dependency. Applications using
Web services can gain performance benefits by reducing the number
of DME service calls; this can be accomplished by using a task with
input objects instead of named references. However, with unnamed
inputs, objects in the task cannot be reused by other tasks.
The JDM Schema maps the JDM matrix objects using a sparse rep-
resentation. In the JDM API, there are three types of matrix objects:
cost matrix, confusion matrix, and similarity matrix . For example, con-
sider the simple cost matrix shown in Figure 10-2. Here the target
attribute has two values: Yes and No . Note that the diagonal elements
for a cost matrix are always zero, because there is no cost associated
with the correct predictions. For nondiagonal elements, the default
cost value is 1. Therefore, to create this cost matrix object, the user
specifies only the nondiagonal element that has nondefault cost
value 3. Listing 10-5(a) shows the XML document for this cost
matrix; it first lists all the target values using the category element and
then specifies the false-negative (FN) cell that has cost value 3.
Because it is known that all diagonal elements have cost value 0 and
unspecified nondiagonal elements have cost value 1, these elements
do not have to be specified in the document.
Cost matrix table.