added L1 regularization and L2 regularization

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

@@ -47,7 +47,8 @@ class LightGBMClassifier(override val uid: String)
     ClassifierTrainParams(getParallelism, getNumIterations, getLearningRate, getNumLeaves,
       getMaxBin, getBaggingFraction, getBaggingFreq, getBaggingSeed, getEarlyStoppingRound,
       getFeatureFraction, getMaxDepth, getMinSumHessianInLeaf, numWorkers, getObjective, modelStr,
-      getIsUnbalance, getVerbosity, categoricalIndexes, actualNumClasses, metric, getBoostFromAverage, getBoostingType)
+      getIsUnbalance, getVerbosity, categoricalIndexes, actualNumClasses, metric, getBoostFromAverage,
+      getBoostingType, getLambdaL1, getLambdaL2)
   }
 
   def getModel(trainParams: TrainParams, lightGBMBooster: LightGBMBooster): LightGBMClassificationModel = {

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

@@ -147,4 +147,16 @@ trait LightGBMParams extends Wrappable with DefaultParamsWritable with HasWeight
 
   def getBoostingType: String = $(boostingType)
   def setBoostingType(value: String): this.type = set(boostingType, value)
+
+  val lambdaL1 = new DoubleParam(this, "lambdaL1", "L1 regularization")
+  setDefault(lambdaL1 -> 0.0)
+
+  def getLambdaL1: Double = $(lambdaL1)
+  def setLambdaL1(value: Double): this.type = set(lambdaL1, value)
+
+  val lambdaL2 = new DoubleParam(this, "lambdaL2", "L2 regularization")
+  setDefault(lambdaL2 -> 0.0)
+
+  def getLambdaL2: Double = $(lambdaL2)
+  def setLambdaL2(value: Double): this.type = set(lambdaL2, value)
 }

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

@@ -58,7 +58,7 @@ class LightGBMRegressor(override val uid: String)
     RegressorTrainParams(getParallelism, getNumIterations, getLearningRate, getNumLeaves,
       getObjective, getAlpha, getTweedieVariancePower, getMaxBin, getBaggingFraction, getBaggingFreq, getBaggingSeed,
       getEarlyStoppingRound, getFeatureFraction, getMaxDepth, getMinSumHessianInLeaf, numWorkers, modelStr,
-      getVerbosity, categoricalIndexes, getBoostFromAverage, getBoostingType)
+      getVerbosity, categoricalIndexes, getBoostFromAverage, getBoostingType, getLambdaL1, getLambdaL2)
   }
 
   def getModel(trainParams: TrainParams, lightGBMBooster: LightGBMBooster): LightGBMRegressionModel = {

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

@@ -25,6 +25,8 @@ abstract class TrainParams extends Serializable {
   def categoricalFeatures: Array[Int]
   def boostFromAverage: Boolean
   def boostingType: String
+  def lambdaL1: Double
+  def lambdaL2: Double
 
   override def toString(): String = {
     s"is_pre_partition=True boosting_type=$boostingType tree_learner=$parallelism num_iterations=$numIterations " +
@@ -33,7 +35,7 @@ abstract class TrainParams extends Serializable {
       s"bagging_seed=$baggingSeed early_stopping_round=$earlyStoppingRound " +
       s"feature_fraction=$featureFraction max_depth=$maxDepth min_sum_hessian_in_leaf=$minSumHessianInLeaf " +
       s"num_machines=$numMachines objective=$objective verbosity=$verbosity " +
-      s"boost_from_average=${boostFromAverage.toString} " +
+      s"boost_from_average=${boostFromAverage.toString} lambda_l1=$lambdaL1 lambda_l2=$lambdaL2 " +
       (if (categoricalFeatures.isEmpty) "" else s"categorical_feature=${categoricalFeatures.mkString(",")}")
   }
 }
@@ -47,7 +49,7 @@ case class ClassifierTrainParams(val parallelism: String, val numIterations: Int
                                  val numMachines: Int, val objective: String, val modelString: Option[String],
                                  val isUnbalance: Boolean, val verbosity: Int, val categoricalFeatures: Array[Int],
                                  val numClass: Int, val metric: String, val boostFromAverage: Boolean,
-                                 val boostingType: String)
+                                 val boostingType: String, val lambdaL1: Double, val lambdaL2: Double)
   extends TrainParams {
   override def toString(): String = {
     val extraStr =
@@ -67,7 +69,7 @@ case class RegressorTrainParams(val parallelism: String, val numIterations: Int,
                                 val maxDepth: Int, val minSumHessianInLeaf: Double, val numMachines: Int,
                                 val modelString: Option[String], val verbosity: Int,
                                 val categoricalFeatures: Array[Int], val boostFromAverage: Boolean,
-                                val boostingType: String)
+                                val boostingType: String, val lambdaL1: Double, val lambdaL2: Double)
   extends TrainParams {
   override def toString(): String = {
     s"alpha=$alpha tweedie_variance_power=$tweedieVariancePower ${super.toString}"

src/lightgbm/src/main/scala/TrainUtils.scala

@@ -269,6 +269,23 @@ private object TrainUtils extends Serializable {
     }.get
   }
 
+  def networkInit(nodes: String, localListenPort: Int, log: Logger, retry: Int, delay: Long): Unit = {
+    try {
+      LightGBMUtils.validate(lightgbmlib.LGBM_NetworkInit(nodes, localListenPort,
+        LightGBMConstants.defaultListenTimeout, nodes.split(",").length), "Network init")
+    } catch {
+      case ex: Throwable => {
+        log.info(s"NetworkInit failed with exception on local port $localListenPort, retrying: $ex")
+        Thread.sleep(delay)
+        if (retry > 0) {
+          networkInit(nodes, localListenPort, log, retry - 1, delay * 2)
+        } else {
+          throw ex
+        }
+      }
+    }
+  }
+
   def trainLightGBM(networkParams: NetworkParams, labelColumn: String, featuresColumn: String,
                     weightColumn: Option[String], validationData: Option[Broadcast[Array[Row]]], log: Logger,
                     trainParams: TrainParams, numCoresPerExec: Int)
@@ -293,8 +310,9 @@ private object TrainUtils extends Serializable {
       // Initialize the network communication
       log.info(s"LightGBM worker listening on: $localListenPort")
       try {
-        LightGBMUtils.validate(lightgbmlib.LGBM_NetworkInit(nodes, localListenPort,
-          LightGBMConstants.defaultListenTimeout, nodes.split(",").length), "Network init")
+        val retries = 3
+        val initialDelay = 1000L
+        networkInit(nodes, localListenPort, log, retries, initialDelay)
         translate(labelColumn, featuresColumn, weightColumn, validationData, log, trainParams, inputRows)
       } finally {
         // Finalize network when done

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

@@ -69,6 +69,8 @@ class VerifyLightGBMClassifier extends Benchmarks with EstimatorFuzzing[LightGBM
     val paramGrid = new ParamGridBuilder()
       .addGrid(lgbm.numLeaves, Array(5, 10))
       .addGrid(lgbm.numIterations, Array(10, 20))
+      .addGrid(lgbm.lambdaL1, Array(0.1, 0.5))
+      .addGrid(lgbm.lambdaL2, Array(0.1, 0.5))
       .build()
 
     val trainValidationSplit = new TrainValidationSplit()
@@ -82,6 +84,10 @@ class VerifyLightGBMClassifier extends Benchmarks with EstimatorFuzzing[LightGBM
     val model = trainValidationSplit.fit(featurizer.transform(dataset))
     model.transform(featurizer.transform(dataset))
     assert(model != null)
+    // Validate lambda parameters set on model
+    val modelStr = model.bestModel.asInstanceOf[LightGBMClassificationModel].getModel.model
+    assert(modelStr.contains("[lambda_l1: 0.1]") || modelStr.contains("[lambda_l1: 0.5]"))
+    assert(modelStr.contains("[lambda_l2: 0.1]") || modelStr.contains("[lambda_l2: 0.5]"))
   }
 
   test("Verify LightGBM Classifier with weight column") {

src/lightgbm/src/test/scala/VerifyLightGBMRegressor.scala

@@ -50,6 +50,8 @@ class VerifyLightGBMRegressor extends Benchmarks with EstimatorFuzzing[LightGBMR
     val paramGrid = new ParamGridBuilder()
       .addGrid(lgbm.numLeaves, Array(5, 10))
       .addGrid(lgbm.numIterations, Array(10, 20))
+      .addGrid(lgbm.lambdaL1, Array(0.1, 0.5))
+      .addGrid(lgbm.lambdaL2, Array(0.1, 0.5))
       .build()
 
     val trainValidationSplit = new TrainValidationSplit()
@@ -63,6 +65,10 @@ class VerifyLightGBMRegressor extends Benchmarks with EstimatorFuzzing[LightGBMR
     val model = trainValidationSplit.fit(featurizer.transform(dataset))
     model.transform(featurizer.transform(dataset))
     assert(model != null)
+    // Validate lambda parameters set on model
+    val modelStr = model.bestModel.asInstanceOf[LightGBMRegressionModel].getModel.model
+    assert(modelStr.contains("[lambda_l1: 0.1]") || modelStr.contains("[lambda_l1: 0.5]"))
+    assert(modelStr.contains("[lambda_l2: 0.1]") || modelStr.contains("[lambda_l2: 0.5]"))
   }
 
   test("Verify LightGBM Regressor with weight column") {