Added class for Mean centering

Author: mandar2812

dynaml-core/src/main/scala-2.11/io/github/mandar2812/dynaml/DynaMLPipe.scala

@@ -28,7 +28,7 @@ import io.github.mandar2812.dynaml.models.sgp.ESGPModel
 import io.github.mandar2812.dynaml.optimization._
 import io.github.mandar2812.dynaml.pipes._
 import io.github.mandar2812.dynaml.probability.ContinuousDistrRV
-import io.github.mandar2812.dynaml.utils.{GaussianScaler, MVGaussianScaler, MinMaxScaler}
+import io.github.mandar2812.dynaml.utils.{GaussianScaler, MVGaussianScaler, MeanScaler, MinMaxScaler}
 import io.github.mandar2812.dynaml.wavelets.{GroupedHaarWaveletFilter, HaarWaveletFilter, InvGroupedHaarWaveletFilter, InverseHaarWaveletFilter}
 import org.apache.log4j.Logger
 import org.apache.spark.rdd.RDD
@@ -383,6 +383,31 @@ object DynaMLPipe {
       (result, (featuresScaler, targetsScaler))
     })
 
+  /**
+    * Returns a pipe which takes a data set and mean centers it.
+    * @param standardize Set to true if one wants the standardized data and false if one
+    *                    does wants the original data with the [[MeanScaler]] instances.
+    * */
+  def calculateMeanScales(standardize: Boolean = true): DataPipe[
+    Stream[(DenseVector[Double], DenseVector[Double])],
+    (Stream[(DenseVector[Double], DenseVector[Double])], (MeanScaler, MeanScaler))] =
+    DataPipe((data: Stream[(DenseVector[Double], DenseVector[Double])]) => {
+
+      val (num_features, num_targets) = (data.head._1.length, data.head._2.length)
+
+      val (mean, _) = utils.getStats(data.map(tup =>
+        DenseVector(tup._1.toArray ++ tup._2.toArray)).toList)
+
+      val featuresScaler = MeanScaler(mean(0 until num_features))
+
+      val targetsScaler = MeanScaler(mean(num_features until num_features + num_targets))
+
+      val result = if(standardize) (featuresScaler * targetsScaler)(data) else data
+
+      (result, (featuresScaler, targetsScaler))
+    })
+
+
   /**
     * Multivariate version of [[calculateGaussianScales]]
     * @param standardize Set to true if one wants the standardized data and false if one

dynaml-core/src/main/scala-2.11/io/github/mandar2812/dynaml/utils/GaussianScaler.scala

@@ -22,19 +22,19 @@ import breeze.linalg.{DenseMatrix, DenseVector, cholesky, inv}
 import io.github.mandar2812.dynaml.pipes.{ReversibleScaler, Scaler}
 
 /**
-  * @author mandar2812 date: 17/6/16.
-  *
   * Scales attributes of a vector pattern using the sample mean and variance of
   * each dimension. This assumes that there is no covariance between the data
   * dimensions.
   *
   * @param mean Sample mean of the data
   * @param sigma Sample variance of each data dimension
-  */
+  * @author mandar2812 date: 17/6/16.
+  *
+  * */
 case class GaussianScaler(mean: DenseVector[Double], sigma: DenseVector[Double])
   extends ReversibleScaler[DenseVector[Double]]{
   override val i: Scaler[DenseVector[Double]] =
-    Scaler((pattern: DenseVector[Double]) => (pattern :* sigma) + mean)
+    Scaler((pattern: DenseVector[Double]) => (pattern *:* sigma) + mean)
 
   override def run(data: DenseVector[Double]): DenseVector[Double] = (data-mean) :/ sigma
 

dynaml-core/src/main/scala-2.11/io/github/mandar2812/dynaml/utils/MeanScaler.scala

@@ -0,0 +1,26 @@
+package io.github.mandar2812.dynaml.utils
+
+import breeze.linalg.DenseVector
+import io.github.mandar2812.dynaml.pipes.{ReversibleScaler, Scaler}
+
+/**
+  * @author mandar date 30/05/2017.
+  * */
+case class MeanScaler(center: DenseVector[Double]) extends ReversibleScaler[DenseVector[Double]] {
+
+  override val i = Scaler((data: DenseVector[Double]) => data + center)
+
+  override def run(data: DenseVector[Double]) = data - center
+
+  def apply(r: Range): MeanScaler = MeanScaler(center(r))
+
+  def apply(n: Int): UnivariateMeanScaler = UnivariateMeanScaler(center(n))
+
+}
+
+case class UnivariateMeanScaler(center: Double) extends ReversibleScaler[Double] {
+
+  override val i = Scaler((data: Double) => data + center)
+
+  override def run(data: Double) = data - center
+}