Update default params, fixed style cop

microsoft · Apr 13, 2018 · a51d736 · a51d736
1 parent b4cc756
commit a51d736
Show file tree

Hide file tree

Showing 3 changed files with 16 additions and 13 deletions.
diff --git a/src/lightgbm/src/main/scala/LightGBMClassifier.scala b/src/lightgbm/src/main/scala/LightGBMClassifier.scala
@@ -41,7 +41,8 @@ class LightGBMClassifier(override val uid: String)
     val encoder = Encoders.kryo[LightGBMBooster]
     log.info(s"Nodes used for LightGBM: $nodes")
     val trainParams = ClassifierTrainParams(getParallelism, getNumIterations, getLearningRate, getNumLeaves,
-      getMaxBin, getBaggingFraction, getBaggingFreq, getBaggingSeed, getFeatureFraction, getMaxDepth, getMinSumHessianInLeaf)
+      getMaxBin, getBaggingFraction, getBaggingFreq, getBaggingSeed, getFeatureFraction,
+      getMaxDepth, getMinSumHessianInLeaf)
     val lightGBMBooster = df
       .mapPartitions(TrainUtils.trainLightGBM(nodes, numNodes, getLabelCol, getFeaturesCol,
         getDefaultListenPort, log, trainParams))(encoder)

diff --git a/src/lightgbm/src/main/scala/LightGBMParams.scala b/src/lightgbm/src/main/scala/LightGBMParams.scala
@@ -36,7 +36,7 @@ trait LightGBMParams extends MMLParams {
   def getNumLeaves: Int = $(numLeaves)
   def setNumLeaves(value: Int): this.type = set(numLeaves, value)
 
-  val maxBin = IntParam(this, "maxBin", "Max bin", 256)
+  val maxBin = IntParam(this, "maxBin", "Max bin", 255)
 
   def getMaxBin: Int = $(maxBin)
   def setMaxBin(value: Int): this.type = set(maxBin, value)
@@ -46,12 +46,12 @@ trait LightGBMParams extends MMLParams {
   def getBaggingFraction: Double = $(baggingFraction)
   def setBaggingFraction(value: Double): this.type = set(baggingFraction, value)
 
-  val baggingFreq = IntParam(this, "baggingFreq", "Bagging frequence", 1)
+  val baggingFreq = IntParam(this, "baggingFreq", "Bagging frequence", 0)
 
   def getBaggingFreq: Int = $(baggingFreq)
   def setBaggingFreq(value: Int): this.type = set(baggingFreq, value)
 
-  val baggingSeed = IntParam(this, "baggingSeed", "Bagging seed", 1)
+  val baggingSeed = IntParam(this, "baggingSeed", "Bagging seed", 3)
 
   def getBaggingSeed: Int = $(baggingSeed)
   def setBaggingSeed(value: Int): this.type = set(baggingSeed, value)
@@ -61,13 +61,13 @@ trait LightGBMParams extends MMLParams {
   def getFeatureFraction: Double = $(featureFraction)
   def setFeatureFraction(value: Double): this.type = set(featureFraction, value)
 
-  val maxDepth = IntParam(this, "maxDepth", "Max depth", 1)
+  val maxDepth = IntParam(this, "maxDepth", "Max depth", -1)
 
   def getMaxDepth: Int = $(maxDepth)
   def setMaxDepth(value: Int): this.type = set(maxDepth, value)
 
-  val minSumHessianInLeaf = IntParam(this, "minSumHessianInLeaf", "minimal sum hessian in one leaf", 1)
+  val minSumHessianInLeaf = DoubleParam(this, "minSumHessianInLeaf", "minimal sum hessian in one leaf", 1e-3)
 
-  def getMinSumHessianInLeaf: Int = $(minSumHessianInLeaf)
-  def setMinSumHessianInLeaf(value: Int): this.type = set(minSumHessianInLeaf, value)
+  def getMinSumHessianInLeaf: Double = $(minSumHessianInLeaf)
+  def setMinSumHessianInLeaf(value: Double): this.type = set(minSumHessianInLeaf, value)
 }
diff --git a/src/lightgbm/src/main/scala/TrainParams.scala b/src/lightgbm/src/main/scala/TrainParams.scala
@@ -16,12 +16,12 @@ abstract class TrainParams extends Serializable {
   def baggingSeed: Int
   def featureFraction: Double
   def maxDepth: Int
-  def minSumHessianInLeaf: Int
+  def minSumHessianInLeaf: Double
 
   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=$baggingFreq bagging_seed=$baggingSeed" +
+      s"learning_rate=$learningRate num_leaves=$numLeaves " +
+      s"max_bin=$maxBin bagging_fraction=$baggingFraction bagging_freq=$baggingFreq bagging_seed=$baggingSeed " +
       s"feature_fraction=$featureFraction max_depth=$maxDepth min_sum_hessian_in_leaf=$minSumHessianInLeaf"
   }
 }
@@ -30,7 +30,8 @@ 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 featureFraction: Double, val maxDepth: Int, val minSumHessianInLeaf: Int)
+                                 val baggingSeed: Int, val featureFraction: Double, val maxDepth: Int,
+                                 val minSumHessianInLeaf: Double)
   extends TrainParams {
   override def toString(): String = {
     s"objective=binary metric=binary_logloss,auc ${super.toString()}"
@@ -42,7 +43,8 @@ case class ClassifierTrainParams(val parallelism: String, val numIterations: Int
 case class RegressorTrainParams(val parallelism: String, val numIterations: Int, val learningRate: Double,
                                 val numLeaves: Int, val application: String, val alpha: Double, val maxBin: Int,
                                 val baggingFraction: Double, val baggingFreq: Int,
-                                val baggingSeed: Int, val featureFraction: Double, val maxDepth: Int, val minSumHessianInLeaf: Int)
+                                val baggingSeed: Int, val featureFraction: Double,
+                                val maxDepth: Int, val minSumHessianInLeaf: Double)
   extends TrainParams {
   override def toString(): String = {
     s"objective=$application alpha=$alpha ${super.toString()}"