Added numClasses and objective, infer actualNumClasses from objective (…

…microsoft#348)

* Added numClasses and objective, infer actualNumClasses from objective

* Update LightGBM notebook example

* Remove numClasses since it is now inferred from dataset

notebooks/samples/106 - Quantile Regression with LightGBM.ipynb

@@ -74,7 +74,7 @@
    "outputs": [],
    "source": [
     "from mmlspark import LightGBMRegressor\n",
-    "model = LightGBMRegressor(application='quantile',\n",
+    "model = LightGBMRegressor(objective='quantile',\n",
     "                          alpha=0.2,\n",
     "                          learningRate=0.3,\n",
     "                          numLeaves=31).fit(train)"

src/lightgbm/src/main/scala/LightGBMBooster.scala

@@ -17,18 +17,6 @@ class LightGBMBooster(val model: String) extends Serializable {
   @transient
   var boosterPtr: SWIGTYPE_p_void = null
 
-  def numClasses(): Int = {
-    if (boosterPtr == null) {
-      LightGBMUtils.initializeNativeLibrary()
-      boosterPtr = getModel()
-    }
-    val numClasses = lightgbmlib.new_intp()
-    LightGBMUtils.validate(
-      lightgbmlib.LGBM_BoosterGetNumClasses(boosterPtr, numClasses),
-      "Booster GetNumClasses")
-    lightgbmlib.intp_value(numClasses)
-  }
-
   def score(features: Vector, raw: Boolean): Double = {
     // Reload booster on each node
     if (boosterPtr == null) {

src/lightgbm/src/main/scala/LightGBMClassifier.scala

@@ -50,7 +50,11 @@ class LightGBMClassifier(override val uid: String)
     log.info(s"Nodes used for LightGBM: ${nodes.mkString(",")}")
     val trainParams = ClassifierTrainParams(getParallelism, getNumIterations, getLearningRate, getNumLeaves,
       getMaxBin, getBaggingFraction, getBaggingFreq, getBaggingSeed, getEarlyStoppingRound,
-      getFeatureFraction, getMaxDepth, getMinSumHessianInLeaf, numWorkers)
+      getFeatureFraction, getMaxDepth, getMinSumHessianInLeaf, numWorkers, getObjective)
+    /* The native code for getting numClasses is always 1 unless it is multiclass-classification problem
+     * so we infer the actual numClasses from the dataset here
+     */
+    val actualNumClasses = getNumClasses(dataset)
     val networkParams = NetworkParams(nodes.toMap, getDefaultListenPort, inetAddress, port)
     val lightGBMBooster = df
       .mapPartitions(TrainUtils.trainLightGBM(networkParams, getLabelCol, getFeaturesCol,
@@ -60,7 +64,7 @@ class LightGBMClassifier(override val uid: String)
     Await.result(future, Duration(getTimeout, SECONDS))
     new LightGBMClassificationModel(uid, lightGBMBooster, getLabelCol, getFeaturesCol,
       getPredictionCol, getProbabilityCol, getRawPredictionCol,
-      if (isDefined(thresholds)) Some(getThresholds) else None)
+      if (isDefined(thresholds)) Some(getThresholds) else None, actualNumClasses)
   }
 
   override def copy(extra: ParamMap): LightGBMClassifier = defaultCopy(extra)
@@ -71,7 +75,8 @@ class LightGBMClassifier(override val uid: String)
 class LightGBMClassificationModel(
   override val uid: String, model: LightGBMBooster, labelColName: String,
   featuresColName: String, predictionColName: String, probColName: String,
-  rawPredictionColName: String, thresholdValues: Option[Array[Double]])
+  rawPredictionColName: String, thresholdValues: Option[Array[Double]],
+  actualNumClasses: Int)
     extends ProbabilisticClassificationModel[Vector, LightGBMClassificationModel]
     with ConstructorWritable[LightGBMClassificationModel] {
 
@@ -96,7 +101,7 @@ class LightGBMClassificationModel(
     }
   }
 
-  override def numClasses: Int = model.numClasses()
+  override def numClasses: Int = this.actualNumClasses
 
   override protected def predictRaw(features: Vector): Vector = {
     val prediction = model.score(features, true)
@@ -105,14 +110,14 @@ class LightGBMClassificationModel(
 
   override def copy(extra: ParamMap): LightGBMClassificationModel =
     new LightGBMClassificationModel(uid, model, labelColName, featuresColName, predictionColName, probColName,
-      rawPredictionColName, thresholdValues)
+      rawPredictionColName, thresholdValues, actualNumClasses)
 
   override val ttag: TypeTag[LightGBMClassificationModel] =
     typeTag[LightGBMClassificationModel]
 
   override def objectsToSave: List[Any] =
     List(uid, model, getLabelCol, getFeaturesCol, getPredictionCol,
-         getProbabilityCol, getRawPredictionCol, thresholdValues)
+         getProbabilityCol, getRawPredictionCol, thresholdValues, actualNumClasses)
 
   def saveNativeModel(session: SparkSession, filename: String): Unit = {
     model.saveNativeModel(session, filename)

src/lightgbm/src/main/scala/LightGBMParams.scala

@@ -36,6 +36,14 @@ trait LightGBMParams extends MMLParams {
   def getNumLeaves: Int = $(numLeaves)
   def setNumLeaves(value: Int): this.type = set(numLeaves, value)
 
+  val objective = StringParam(this, "objective",
+    "The Objective. For regression applications, this can be: " +
+    "regression_l2, regression_l1, huber, fair, poisson, quantile, mape, gamma or tweedie. " +
+    "For classification applications, this can be: binary, multiclass, or multiclassova. ", "regression")
+
+  def getObjective: String = $(objective)
+  def setObjective(value: String): this.type = set(objective, value)
+
   val maxBin = IntParam(this, "maxBin", "Max bin", 255)
 
   def getMaxBin: Int = $(maxBin)

src/lightgbm/src/main/scala/LightGBMRegressor.scala

@@ -44,13 +44,6 @@ class LightGBMRegressor(override val uid: String)
   def getAlpha: Double = $(alpha)
   def setAlpha(value: Double): this.type = set(alpha, value)
 
-  val application = StringParam(this, "application",
-    "Regression application, regression_l2, regression_l1, huber, fair, poisson, quantile, mape, gamma or tweedie",
-    "regression")
-
-  def getApplication: String = $(application)
-  def setApplication(value: String): this.type = set(application, value)
-
   /** Trains the LightGBM Regression model.
     *
     * @param dataset The input dataset to train.
@@ -72,7 +65,7 @@ class LightGBMRegressor(override val uid: String)
     val encoder = Encoders.kryo[LightGBMBooster]
     log.info(s"Nodes used for LightGBM: ${nodes.mkString(",")}")
     val trainParams = RegressorTrainParams(getParallelism, getNumIterations, getLearningRate, getNumLeaves,
-      getApplication, getAlpha, getMaxBin, getBaggingFraction, getBaggingFreq, getBaggingSeed, getEarlyStoppingRound,
+      getObjective, getAlpha, getMaxBin, getBaggingFraction, getBaggingFreq, getBaggingSeed, getEarlyStoppingRound,
       getFeatureFraction, getMaxDepth, getMinSumHessianInLeaf, numWorkers)
     val networkParams = NetworkParams(nodes.toMap, getDefaultListenPort, inetAddress, port)
     val lightGBMBooster = df

src/lightgbm/src/main/scala/TrainParams.scala

@@ -19,14 +19,15 @@ abstract class TrainParams extends Serializable {
   def maxDepth: Int
   def minSumHessianInLeaf: Double
   def numMachines: Int
+  def objective: String
 
   override def toString(): String = {
     s"is_pre_partition=True boosting_type=gbdt tree_learner=$parallelism num_iterations=$numIterations " +
       s"learning_rate=$learningRate num_leaves=$numLeaves " +
       s"max_bin=$maxBin bagging_fraction=$baggingFraction bagging_freq=$baggingFreq " +
       s"bagging_seed=$baggingSeed early_stopping_round=$earlyStoppingRound " +
       s"feature_fraction=$featureFraction max_depth=$maxDepth min_sum_hessian_in_leaf=$minSumHessianInLeaf " +
-      s"num_machines=$numMachines"
+      s"num_machines=$numMachines objective=$objective"
   }
 }
 
@@ -35,22 +36,23 @@ abstract class TrainParams extends Serializable {
 case class ClassifierTrainParams(val parallelism: String, val numIterations: Int, val learningRate: Double,
                                  val numLeaves: Int, val maxBin: Int, val baggingFraction: Double, val baggingFreq: Int,
                                  val baggingSeed: Int, val earlyStoppingRound: Int, val featureFraction: Double,
-                                 val maxDepth: Int, val minSumHessianInLeaf: Double, val numMachines: Int)
+                                 val maxDepth: Int, val minSumHessianInLeaf: Double,
+                                 val numMachines: Int, val objective: String)
   extends TrainParams {
   override def toString(): String = {
-    s"objective=binary metric=binary_logloss,auc ${super.toString()}"
+    s"metric=binary_logloss,auc ${super.toString()}"
   }
 }
 
 /** Defines the Booster parameters passed to the LightGBM regressor.
   */
 case class RegressorTrainParams(val parallelism: String, val numIterations: Int, val learningRate: Double,
-                                val numLeaves: Int, val application: String, val alpha: Double, val maxBin: Int,
+                                val numLeaves: Int, val objective: String, val alpha: Double, val maxBin: Int,
                                 val baggingFraction: Double, val baggingFreq: Int,
                                 val baggingSeed: Int, val earlyStoppingRound: Int, val featureFraction: Double,
                                 val maxDepth: Int, val minSumHessianInLeaf: Double, val numMachines: Int)
   extends TrainParams {
   override def toString(): String = {
-    s"objective=$application alpha=$alpha ${super.toString()}"
+    s"alpha=$alpha ${super.toString()}"
   }
 }

src/lightgbm/src/test/scala/VerifyLightGBMClassifier.scala

@@ -14,6 +14,7 @@ class VerifyLightGBMClassifier extends Benchmarks with EstimatorFuzzing[LightGBM
   lazy val moduleName = "lightgbm"
   var portIndex = 30
   val numPartitions = 2
+  val objective = "binary"
 
   // TODO: Need to add multiclass param with objective function
   // verifyLearnerOnMulticlassCsvFile("abalone.csv",                  "Rings", 2)
@@ -51,6 +52,7 @@ class VerifyLightGBMClassifier extends Benchmarks with EstimatorFuzzing[LightGBM
       .setDefaultListenPort(LightGBMConstants.defaultLocalListenPort + portIndex)
       .setNumLeaves(5)
       .setNumIterations(10)
+      .setObjective(objective)
 
     val paramGrid = new ParamGridBuilder()
       .addGrid(lgbm.numLeaves, Array(5, 10))
@@ -102,6 +104,7 @@ class VerifyLightGBMClassifier extends Benchmarks with EstimatorFuzzing[LightGBM
         .setDefaultListenPort(LightGBMConstants.defaultLocalListenPort + portIndex)
         .setNumLeaves(5)
         .setNumIterations(10)
+        .setObjective(objective)
         .fit(trainData)
       val scoredResult = model.transform(trainData).drop(featuresColumn)
       val splitFeatureImportances = model.getFeatureImportances("split")
@@ -134,6 +137,7 @@ class VerifyLightGBMClassifier extends Benchmarks with EstimatorFuzzing[LightGBM
         .setDefaultListenPort(LightGBMConstants.defaultLocalListenPort + portIndex)
         .setNumLeaves(5)
         .setNumIterations(10)
+        .setObjective(objective)
         .fit(trainData)
       val scoredResult = model.transform(trainData).drop(featuresColumn)
       val splitFeatureImportances = model.getFeatureImportances("split")
@@ -161,7 +165,8 @@ class VerifyLightGBMClassifier extends Benchmarks with EstimatorFuzzing[LightGBM
       new LightGBMClassifier()
         .setLabelCol(labelCol)
         .setFeaturesCol(featuresCol)
-        .setNumLeaves(5),
+        .setNumLeaves(5)
+        .setObjective(objective),
       train))
   }
 
@@ -183,6 +188,7 @@ class VerifyLightGBMClassifier extends Benchmarks with EstimatorFuzzing[LightGBM
         .setDefaultListenPort(LightGBMConstants.defaultLocalListenPort + portIndex)
         .setNumLeaves(5)
         .setNumIterations(10)
+        .setObjective(objective)
         .fit(featurizer.transform(dataset))
 
       val targetDir: Path = Paths.get(getClass.getResource("/").toURI)