lazy val on other aggregators

Author: sethah

mllib/src/main/scala/org/apache/spark/ml/classification/LinearSVC.scala

@@ -458,6 +458,7 @@ private class LinearSVCAggregator(
    */
   def add(instance: Instance): this.type = {
     instance match { case Instance(label, weight, features) =>
+      require(weight >= 0.0, s"instance weight, $weight has to be >= 0.0")
       require(numFeatures == features.size, s"Dimensions mismatch when adding new instance." +
         s" Expecting $numFeatures but got ${features.size}.")
       if (weight == 0.0) return this

mllib/src/main/scala/org/apache/spark/ml/classification/LogisticRegression.scala

@@ -1436,7 +1436,7 @@ private class LogisticAggregator(
     case _ => throw new IllegalArgumentException(s"coefficients only supports dense vector but " +
       s"got type ${bcCoefficients.value.getClass}.)")
   }
-  private val gradientSumArray = new Array[Double](coefficientSize)
+  private lazy val gradientSumArray = new Array[Double](coefficientSize)
 
   if (multinomial && numClasses <= 2) {
     logInfo(s"Multinomial logistic regression for binary classification yields separate " +

mllib/src/main/scala/org/apache/spark/ml/clustering/GaussianMixture.scala

@@ -580,10 +580,10 @@ private class ExpectationAggregator(
   private val k: Int = bcWeights.value.length
   private var totalCnt: Long = 0L
   private var newLogLikelihood: Double = 0.0
-  private val newWeights: Array[Double] = new Array[Double](k)
-  private val newMeans: Array[DenseVector] = Array.fill(k)(
+  private lazy val newWeights: Array[Double] = new Array[Double](k)
+  private lazy val newMeans: Array[DenseVector] = Array.fill(k)(
     new DenseVector(Array.fill[Double](numFeatures)(0.0)))
-  private val newCovs: Array[DenseVector] = Array.fill(k)(
+  private lazy val newCovs: Array[DenseVector] = Array.fill(k)(
     new DenseVector(Array.fill[Double](numFeatures * (numFeatures + 1) / 2)(0.0)))
 
   @transient private lazy val oldGaussians = {

mllib/src/main/scala/org/apache/spark/ml/regression/AFTSurvivalRegression.scala

@@ -526,7 +526,7 @@ private class AFTAggregator(
   private var totalCnt: Long = 0L
   private var lossSum = 0.0
   // Here we optimize loss function over log(sigma), intercept and coefficients
-  private val gradientSumArray = Array.ofDim[Double](length)
+  private lazy val gradientSumArray = Array.ofDim[Double](length)
 
   def count: Long = totalCnt
   def loss: Double = {

mllib/src/main/scala/org/apache/spark/ml/regression/LinearRegression.scala

@@ -959,7 +959,7 @@ private class LeastSquaresAggregator(
   @transient private lazy val effectiveCoefficientsVector = effectiveCoefAndOffset._1
   @transient private lazy val offset = effectiveCoefAndOffset._2
 
-  private val gradientSumArray = Array.ofDim[Double](dim)
+  private lazy val gradientSumArray = Array.ofDim[Double](dim)
 
   /**
    * Add a new training instance to this LeastSquaresAggregator, and update the loss and gradient