Builder property for strict input schema check

[PsychedelicO0o]

Hi：
I've trained a pipeline model in SparkML and converted it to PMML. When I load it by using LoadingModelEvaluatorBuilder() in Java and pass a feature map into it to get result, it throws an Exception:

Exception in thread "main" org.jpmml.evaluator.MissingFieldException: Field "pmml(prediction)" is not defined
	at org.jpmml.evaluator.ModelEvaluationContext.resolve(ModelEvaluationContext.java:149)
	at org.jpmml.evaluator.EvaluationContext.evaluate(EvaluationContext.java:95)
	at org.jpmml.evaluator.ExpressionUtil.evaluateMapValues(ExpressionUtil.java:321)
	at org.jpmml.evaluator.ExpressionUtil.evaluateExpression(ExpressionUtil.java:186)
	at org.jpmml.evaluator.ExpressionUtil.evaluate(ExpressionUtil.java:156)
	at org.jpmml.evaluator.ExpressionUtil.evaluateExpressionContainer(ExpressionUtil.java:82)
	at org.jpmml.evaluator.OutputUtil.evaluate(OutputUtil.java:250)
	at org.jpmml.evaluator.ModelEvaluator.evaluateInternal(ModelEvaluator.java:470)
	at org.jpmml.evaluator.mining.MiningModelEvaluator.evaluateInternal(MiningModelEvaluator.java:237)
	at org.jpmml.evaluator.ModelEvaluator.evaluate(ModelEvaluator.java:302)
	at com.jd.campus.algorithm.pmml.NewTest.main(NewTest.java:35)

Could you please help me find how to solve it and what's the reason, cause some models throw this Exception but others are normal. Thanks
Here Is PMML File And Java Code
JMMPL-Evaluator Version : 1.5.16

[vruusmann]

When I try to run the example project then I get a different exception:

Exception in thread "main" org.jpmml.evaluator.InvalidResultException: Field "quota" cannot accept user input value 100.0
        at org.jpmml.evaluator.InputFieldUtil.performInvalidValueTreatment(InputFieldUtil.java:252)
        at org.jpmml.evaluator.InputFieldUtil.prepareScalarInputValue(InputFieldUtil.java:151)
        at org.jpmml.evaluator.InputFieldUtil.prepareInputValue(InputFieldUtil.java:111)
        at org.jpmml.evaluator.ModelEvaluationContext.prepare(ModelEvaluationContext.java:76)

[vruusmann]

Anyway, the root cause of the original exception (a MissingFieldException) is that some of the input fields (aka features) have missing values, the first stage of the model then predicts a missing value, and this first stage prediction cannot be post-processed by the Output/OutputField element.

This issue is a functional duplicate of #237

I did rename some relevant exception classes in the JPMML-Evalautor 1.6.X development branch to make the troubleshooting process easier.

[vruusmann]

In the current case, the missing input field is red_pack_face_value.

See for yourself:

HashMap<String, Object> featureMap = JSON.parseObject(featureStr, new TypeReference<HashMap<String, Object>>() {});
String pmmlPath = "model_jy.pmml";
Evaluator evaluator = new LoadingModelEvaluatorBuilder().load(new File(pmmlPath)).build();
evaluator.verify();
Set<String> inputFieldNames = evaluator.getInputFields().stream()
	.map(inputField -> inputField.getName().getValue())
	.collect(Collectors.toSet());
inputFieldNames.removeAll(featureMap.keySet());
System.out.println("Missing input fields: " + inputFieldNames);

[vruusmann]

I renamed/re-purposed this issue in order to design a ModelEvaluatorBuilder property that would equip the returned Evaluator instance with extra input fields checking logic.

If this property is activated, then the Evaluator#evaluate(Map) method would check that all input fields are part of the arguments map. If not, the evaluate method would fail fast, without executing the real model business logic at all.

This problem is all too common with end users.