[SPARK-12811][ML] Estimator for Generalized Linear Models(GLMs)

Estimator for Generalized Linear Models(GLMs) which will be solved by IRLS.

cc mengxr

Author: Yanbo Liang <ybliang8@gmail.com>

Closes apache#11136 from yanboliang/spark-12811.

mllib/src/main/scala/org/apache/spark/ml/optim/WeightedLeastSquares.scala

@@ -156,6 +156,12 @@ private[ml] class WeightedLeastSquares(
 
 private[ml] object WeightedLeastSquares {
 
+  /**
+   * In order to take the normal equation approach efficiently, [[WeightedLeastSquares]]
+   * only supports the number of features is no more than 4096.
+   */
+  val MAX_NUM_FEATURES: Int = 4096
+
   /**
    * Aggregator to provide necessary summary statistics for solving [[WeightedLeastSquares]].
    */
@@ -174,8 +180,8 @@ private[ml] object WeightedLeastSquares {
     private var aaSum: DenseVector = _
 
     private def init(k: Int): Unit = {
-      require(k <= 4096, "In order to take the normal equation approach efficiently, " +
-        s"we set the max number of features to 4096 but got $k.")
+      require(k <= MAX_NUM_FEATURES, "In order to take the normal equation approach efficiently, " +
+        s"we set the max number of features to $MAX_NUM_FEATURES but got $k.")
       this.k = k
       triK = k * (k + 1) / 2
       count = 0L