use axpy in Updater

apache · Mar 27, 2014 · e981396 · e981396
1 parent db808a1
commit e981396
Show file tree

Hide file tree

Showing 2 changed files with 21 additions and 10 deletions.
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/optimization/GradientDescent.scala b/mllib/src/main/scala/org/apache/spark/mllib/optimization/GradientDescent.scala
@@ -19,6 +19,8 @@ package org.apache.spark.mllib.optimization
 
 import scala.collection.mutable.ArrayBuffer
 
+import breeze.linalg.{Vector => BV, DenseVector => BDV}
+
 import org.apache.spark.Logging
 import org.apache.spark.rdd.RDD
 import org.apache.spark.mllib.linalg.{Vectors, Vector}
@@ -157,11 +159,16 @@ object GradientDescent extends Logging {
     for (i <- 1 to numIterations) {
       // Sample a subset (fraction miniBatchFraction) of the total data
       // compute and sum up the subgradients on this subset (this is one map-reduce)
-      val (gradientSum, lossSum) = data.sample(false, miniBatchFraction, 42 + i).map {
-        case (y, features) =>
-          val (grad, loss) = gradient.compute(features, y, weights)
-          (grad.toBreeze, loss)
-      }.reduce((a, b) => (a._1 += b._1, a._2 + b._2))
+      val (gradientSum, lossSum) = data.sample(false, miniBatchFraction, 42 + i)
+        .aggregate((BDV.zeros[Double](weights.size), 0.0))(
+          seqOp = (c, v) => (c, v) match { case ((grad, loss), (label, features)) =>
+            val (g, l) = gradient.compute(features, label, weights)
+            (grad += g.toBreeze, loss + l)
+          },
+          combOp = (c1, c2) => (c1, c2) match { case ((grad1, loss1), (grad2, loss2)) =>
+            (grad1 += grad2, loss1 + loss2)
+          }
+        )
 
       /**
        * NOTE(Xinghao): lossSum is computed using the weights from the previous iteration

diff --git a/mllib/src/main/scala/org/apache/spark/mllib/optimization/Updater.scala b/mllib/src/main/scala/org/apache/spark/mllib/optimization/Updater.scala
@@ -19,7 +19,7 @@ package org.apache.spark.mllib.optimization
 
 import scala.math._
 
-import breeze.linalg.{norm => brzNorm}
+import breeze.linalg.{norm => brzNorm, axpy => brzAxpy, Vector => BV}
 
 import org.apache.spark.mllib.linalg.{Vectors, Vector}
 
@@ -70,7 +70,9 @@ class SimpleUpdater extends Updater {
       iter: Int,
       regParam: Double): (Vector, Double) = {
     val thisIterStepSize = stepSize / math.sqrt(iter)
-    val brzWeights = weightsOld.toBreeze - gradient.toBreeze * thisIterStepSize
+    val brzWeights: BV[Double] = weightsOld.toBreeze.toDenseVector
+    brzAxpy(-thisIterStepSize, gradient.toBreeze, brzWeights)
+
     (Vectors.fromBreeze(brzWeights), 0)
   }
 }
@@ -102,7 +104,8 @@ class L1Updater extends Updater {
       regParam: Double): (Vector, Double) = {
     val thisIterStepSize = stepSize / math.sqrt(iter)
     // Take gradient step
-    val brzWeights = weightsOld.toBreeze - gradient.toBreeze * thisIterStepSize
+    val brzWeights: BV[Double] = weightsOld.toBreeze.toDenseVector
+    brzAxpy(-thisIterStepSize, gradient.toBreeze, brzWeights)
     // Apply proximal operator (soft thresholding)
     val shrinkageVal = regParam * thisIterStepSize
     var i = 0
@@ -133,8 +136,9 @@ class SquaredL2Updater extends Updater {
     // w' = w - thisIterStepSize * (gradient + regParam * w)
     // w' = (1 - thisIterStepSize * regParam) * w - thisIterStepSize * gradient
     val thisIterStepSize = stepSize / math.sqrt(iter)
-    val brzWeights = weightsOld.toBreeze * (1.0 - thisIterStepSize * regParam) -
-      (gradient.toBreeze * thisIterStepSize)
+    val brzWeights: BV[Double] = weightsOld.toBreeze.toDenseVector
+    brzWeights :*= (1.0 - thisIterStepSize * regParam)
+    brzAxpy(-thisIterStepSize, gradient.toBreeze, brzWeights)
     val norm = brzNorm(brzWeights, 2.0)
 
     (Vectors.fromBreeze(brzWeights), 0.5 * regParam * norm * norm)