org.apache.spark.mllib.optimization.LogisticGradient Scala Examples

package keystoneml.nodes.learning

import breeze.linalg.Vector
import org.apache.spark.mllib.classification.{LogisticRegressionModel => MLlibLRM}
import org.apache.spark.mllib.linalg.{Vector => MLlibVector}
import org.apache.spark.mllib.optimization.{SquaredL2Updater, LogisticGradient, LBFGS}
import org.apache.spark.mllib.regression.{GeneralizedLinearAlgorithm, LabeledPoint}
import org.apache.spark.mllib.util.DataValidators
import org.apache.spark.rdd.RDD
import keystoneml.utils.MLlibUtils.breezeVectorToMLlib
import keystoneml.workflow.{LabelEstimator, Transformer}

import scala.reflect.ClassTag


  private[this] class LogisticRegressionWithLBFGS(numClasses: Int, numFeaturesValue: Int)
      extends GeneralizedLinearAlgorithm[MLlibLRM] with Serializable {

    this.numFeatures = numFeaturesValue
    override val optimizer = new LBFGS(new LogisticGradient, new SquaredL2Updater)

    override protected val validators = List(multiLabelValidator)

    require(numClasses > 1)
    numOfLinearPredictor = numClasses - 1
    if (numClasses > 2) {
      optimizer.setGradient(new LogisticGradient(numClasses))
    }

    private def multiLabelValidator: RDD[LabeledPoint] => Boolean = { data =>
      if (numOfLinearPredictor > 1) {
        DataValidators.multiLabelValidator(numOfLinearPredictor + 1)(data)
      } else {
        DataValidators.binaryLabelValidator(data)
      }
    }

    override protected def createModel(weights: MLlibVector, intercept: Double) = {
      if (numOfLinearPredictor == 1) {
        new MLlibLRM(weights, intercept)
      } else {
        new MLlibLRM(weights, intercept, numFeatures, numOfLinearPredictor + 1)
      }
    }
  }

  override def fit(in: RDD[T], labels: RDD[Int]): LogisticRegressionModel[T] = {
    val labeledPoints = labels.zip(in).map(x => LabeledPoint(x._1, breezeVectorToMLlib(x._2)))
    val trainer = new LogisticRegressionWithLBFGS(numClasses, numFeatures)
    trainer.setValidateData(false).optimizer.setNumIterations(numIters).setRegParam(regParam)
    val model = trainer.run(labeledPoints)

    new LogisticRegressionModel(model)
  }
} 

package se.uu.farmbio.cp.alg

import org.apache.spark.mllib.classification.LogisticRegressionModel
import org.apache.spark.mllib.linalg.Vector
import org.apache.spark.mllib.linalg.Vectors
import org.apache.spark.mllib.optimization.LBFGS
import org.apache.spark.mllib.optimization.LogisticGradient
import org.apache.spark.mllib.optimization.SquaredL2Updater
import org.apache.spark.mllib.regression.LabeledPoint
import org.apache.spark.mllib.util.MLUtils
import org.apache.spark.rdd.RDD

import se.uu.farmbio.cp.UnderlyingAlgorithm

//Define a LogisticRegression UnderlyingAlgorithm
private object LogisticRegression {
  def trainingProcedure(
    input: RDD[LabeledPoint],
    maxNumItearations: Int,
    regParam: Double,
    numCorrections: Int,
    convergenceTol: Double): (Vector => Double) = {

    //Train Logistic Regression with LBFGS
    val numFeatures = input.take(1)(0).features.size
    val training = input.map(x => (x.label, MLUtils.appendBias(x.features))).cache()
    val initialWeightsWithIntercept = Vectors.dense(new Array[Double](numFeatures + 1))
    val (weightsWithIntercept, _) = LBFGS.runLBFGS(
      training,
      new LogisticGradient(),
      new SquaredL2Updater(),
      numCorrections,
      convergenceTol,
      maxNumItearations,
      regParam,
      initialWeightsWithIntercept)

    //Create the model using the weights
    val model = new LogisticRegressionModel(
      Vectors.dense(weightsWithIntercept.toArray.slice(0, weightsWithIntercept.size - 1)),
      weightsWithIntercept(weightsWithIntercept.size - 1))

    //Return raw score predictor
    model.clearThreshold()
    model.predict

  }
}

class LogisticRegression(
  private val input: RDD[LabeledPoint],
  private val maxNumItearations: Int = 100,
  private val regParam: Double = 0.1,
  private val numCorrections: Int = 10,
  private val convergenceTol: Double = 1e-4)
  extends UnderlyingAlgorithm(
    LogisticRegression.trainingProcedure(
      input,
      maxNumItearations,
      regParam,
      numCorrections,
      convergenceTol)) {
  override def nonConformityMeasure(newSample: LabeledPoint) = {
    val score = predictor(newSample.features)
    if (newSample.label == 1.0) {
      1-score
    } else {
      score
    }
  }
} 

package linalg.sgd
import scala.util.Random
import org.apache.spark.mllib.linalg.Vectors
import org.apache.spark.mllib.optimization.GradientDescent
import org.apache.spark.mllib.optimization.SquaredL2Updater
import org.apache.spark.mllib.optimization.LogisticGradient
import org.apache.spark.SparkContext



object SparkSGD {
  def main(args: Array[String]): Unit = {
    val m = 4
    val n = 200000
    val sc = new SparkContext("local[2]", "")
    val points = sc.parallelize(0 until m, 2).mapPartitionsWithIndex { (idx, iter) =>
      val random = new Random(idx)
      iter.map(i => (1.0, Vectors.dense(Array.fill(n)(random.nextDouble()))))
    }.cache()
    val (weights, loss) = GradientDescent.runMiniBatchSGD(
      points,
      new LogisticGradient,
      new SquaredL2Updater,
      0.1,
      2,
      1.0,
      1.0,
      Vectors.dense(new Array[Double](n)))
    println("w:"  + weights(0))
    println("loss:" + loss(0))
    sc.stop()

  }
}