The specification has a high number of properties where the range is Concept or Literal. These lead to arbitray uses and create issues regarding interoperability. For example, MachineLearningModel has a property evaluationResults with range Literal. This has two significant issues:
- We need to interpret this information manually, whereas an enumerated taxonomy would provide important context such as
EvaluatedHighErrors or even provide a set of evaluations run PassedTestA -- where both these as instances of some class such as EvaluationResult. This class then allows creating specific instances and subclasses so that evaluations can be structured and compared e.g. TestA is a more specific form of TestB such that passing A also means you have passed B. This is sem-web best practice and a cornerstone of semantic interoperability.
- If there are multiple evaluations run, and we only have literals to use, then the two properties
evaluationResults and evaluationStrategy will have ambiguous combinations. E.g. for these triples :evaluationResults ex:R1, ex:R2 and :evaluationStrategy ex:S1, ex:S2, we cannot state that ex:S1 was used to get ex:R1 and ex:S2 for ex:R2. The solution (as with no.1 above) is to have a class such as Evaluation whose instances can then contain the results, strategies, and also allow additional information such as who conducted this, when it was conducted, etc.
This also occurs in other properties such as intended use, limitations, how to use, etc.
The specification has a high number of properties where the range is
ConceptorLiteral. These lead to arbitray uses and create issues regarding interoperability. For example,MachineLearningModelhas a propertyevaluationResultswith rangeLiteral. This has two significant issues:EvaluatedHighErrorsor even provide a set of evaluations runPassedTestA-- where both these as instances of some class such asEvaluationResult. This class then allows creating specific instances and subclasses so that evaluations can be structured and compared e.g.TestAis a more specific form ofTestBsuch that passing A also means you have passed B. This is sem-web best practice and a cornerstone of semantic interoperability.evaluationResultsandevaluationStrategywill have ambiguous combinations. E.g. for these triples:evaluationResults ex:R1, ex:R2and:evaluationStrategy ex:S1, ex:S2, we cannot state thatex:S1was used to getex:R1andex:S2forex:R2. The solution (as with no.1 above) is to have a class such asEvaluationwhose instances can then contain the results, strategies, and also allow additional information such as who conducted this, when it was conducted, etc.This also occurs in other properties such as intended use, limitations, how to use, etc.