SPP Tool Chain
2.3 Conditional Classifier

SPML supports explicit representation of the decision logic implemented in the signal classifier. This can then be used to automatically synthesize the classifier code.

The modeling elements that are used in representing a dataflow graph are described below. The description consists of a name of the modeling element, the stereotype of the modeling element which indicates the concrete syntactic realization of the element in GME (i.e. Folder, Model, Atom, Connection, Set, etc.), an explanation of the role and purpose of modeling element, and a description of the attributes of the modeling element. For additional details on understanding the concrete syntactic constructs of GME, please refer to the GME documentation.

1.    Classifier

Stereotype:	Model
Description:	Classifier model represents a conditional classifier component. It is derived from a Primitive since a classifier is a primitive signal processing component, with the difference that its implementation is modeled with decision logic. The modeling of Classifier component is composed of the following four aspects:          ▪   AppDataflow  - define the structure of the component          ▪   TypeAspect    - define the datatype of the ports          ▪   ImplAspect     - define the implementation the computational cores          ▪   Classifier         - define the implementation
Attribute:	Name DataType Note ResourceType Enumeration:       CPU, FPGA, SIM,       AD, DA, MEM Specify the resource type of this conditional classifier component.

2.    Condition

Icon:
Stereotype:	Atom
Description:	Condition atoms are used to define the decision logic. The 'Expression' attribute of the Condition object captures a relational expression over the values of a feature which is input to the classifier thru its InputPorts. All the features that are used in an expression are referred to by the InputPort name, and this dependency of the condition on features is also explicitly represented with the PCConn connection. The outcome of the condition is a Boolean value.
Attribute:	Name DataType Note Expression String Captures a relational expression over the values of a feature which is input to the classifier through its InputPorts.

3.    Classification

Icon:
Stereotype:	Atom
Description:	Classification atoms are used to define the decision logic. The 'Expression' attribute of the Classification object captures a Boolean formula over the conditions. This expression refers to Conditions with names, however explicit dependencies are captured by the CDConn connection. Multiple Classification objects capture the classification logic for different signal types. Often in the classification logic redundant or mutually dependent conditions are defined to simplify the decision logic. This information is useful in simplifying the decision logic when synthesizing the conditional classifier.
Attribute:	Name DataType Note Expression String Captures a Boolean formula over the conditions. ClassName String   Probability Double the likelihood of a particular Classification decision FalseAlarmCost Double the cost of falsely recoginizing it MissCost Double the cost of missing it

3.    CondRelation

Stereotype:	Connection
Description:	CondRelation is an association to define the explicit dependencies between Condition objects.
Attribute:	Name DataType Note RelationSematics Enumeration:       Implies,       Implies_Not       Equivalent,       Equivalent_Not Defines the semantics of the relation if it is an implication, equivalence, inverse implication, or inverse equivalence relation.

4.    CDConn

Stereotype:	Connection
Description:	CDConn is an association to define the explicit dependencies between Condition object and Classification object.

5.    PCConn

Stereotype:	Connection
Description:	PCConn is an association to define the explicit dependencies between Condition object and InputPort. It represents that all features used in an expression of the Condition object are referred to by the InputPort name..

The metamodel of the above elements in the dataflow graph is shown as below:

Next Up: 2.4 Datatypes