//设置package包名称以及导入依赖的类
package com.inneractive.cerberus

import java.io.{File, FileInputStream}
import java.util
import javax.xml.transform.stream.StreamSource

import org.dmg.pmml.{FieldName, Model, PMML}
import org.jpmml.evaluator.{EvaluatorUtil, FieldValue, ModelEvaluator, ModelEvaluatorFactory}
import org.jpmml.model.JAXBUtil

import scala.collection.Map
import scala.util.Try



class LRPmmlEvaluator(modelLocation: String, threshold: Option[Double]) {

  val inputStream = new FileInputStream(new File(modelLocation, "/model/PMML/PMML.xml"))
  val pmmModel: PMML = JAXBUtil.unmarshalPMML(new StreamSource(inputStream))

  val evaluatorFactory: ModelEvaluatorFactory = ModelEvaluatorFactory.newInstance()
  val evaluator: ModelEvaluator[_ <: Model] = evaluatorFactory.newModelManager(pmmModel)
  val activeFields: util.List[FieldName] = evaluator.getActiveFields
  val targetFields: util.List[FieldName] = EvaluatorUtil.getTargetFields(evaluator)
  val outputFields: util.List[FieldName] = EvaluatorUtil.getOutputFields(evaluator)

  def arguments(eval: FieldName => (FieldName, FieldValue)) = (activeFields map eval).toMap

  def getResults(result: java.util.Map[FieldName, _]) = {
    val targets = targetFields map (f => EvaluatorUtil.decode(result.get(f)))
    val outputs = outputFields map (f => result.get(f))
    Tuple3(targets.head.asInstanceOf[Double], outputs.head.asInstanceOf[Double], outputs(1).asInstanceOf[Double])
  }

  def callEvaluator(params : Map[FieldName, _]) = getResults(evaluator.evaluate(params))

  def predict(features: Features) = Try {
    val r = callEvaluator(arguments((f: FieldName) => {
      val value = features.vector(f.getValue)
      val activeValue = evaluator.prepare(f, value)
      (f, activeValue)
    }))
    new Prediction("", if (r._1 == 0.0) true else false, if (r._1 == 0) r._2 else r._3)
  }
}