edu.knowitall.tool.conf.BreezeLogisticRegressionTrainer.scala Maven / Gradle / Ivy

Go to download

Show more of this group Show more artifacts with this name
Show all versions of nlptools-conf-breeze_2.9.3 Show documentation

nlptools-conf-breeze

The newest version!

package edu.knowitall
package tool
package conf

import breeze.classify.LogisticClassifier
import breeze.data.Example
import breeze.linalg.DenseVector
import breeze.optimize.FirstOrderMinimizer.OptParams
import edu.knowitall.tool.conf.impl.LogisticRegression

class BreezeLogisticRegressionTrainer[E](features: FeatureSet[E, Double]) extends ConfidenceTrainer[E](features) {
  def trainBreezeClassifier(instances: Iterable[Labelled[E]], optParams: OptParams) = {
    val examples = instances.zipWithIndex map { case (Labelled(label, item: Any), i) =>
      val vector = DenseVector((1.0 +: features.vectorize(item.asInstanceOf[E])).toArray)
      Example[Boolean, DenseVector[Double]](label, vector, id=i.toString)
    }

    new LogisticClassifier.Trainer[Boolean,DenseVector[Double]](optParams).train(examples)
  }

  def train(labelled: Iterable[Labelled[E]], optParams: OptParams): LogisticRegression[E] = {
    val classifier = trainBreezeClassifier(labelled, optParams)

    val weights = (("Intercept" +: features.featureNames).iterator zip classifier.featureWeights.indexed(true).iterator.map(_._2)).toMap
    new LogisticRegression(features, weights, 0.0)
  }

  override def train(labelled: Iterable[Labelled[E]]): LogisticRegression[E] = {
    train(labelled, OptParams(useL1 = true))
  }
}