package dk.nezbo.ir.ass2

import scala.collection.immutable.Set
import ch.ethz.dal.classifier.processing.XMLDocument
import scala.util.Random
import scala.collection.mutable.HashMap

class LogisticRegression(topic: String, trainSet: Seq[(Set[String],Map[String,Double])]) extends Classifier {

  // FIELDS

  val bias = 1.0 // TODO: What it should be?
  val random = new Random(1337)

  var theta: Map[String,Double] = Map[String,Double]()

  // INHERITED FUNCTIONS

  def getTopic() : String = {
    topic
  }

  def train(iteration: Int) : Unit = {
    val lRate = 1.0/iteration.toDouble

    for(doc <- trainSet){
        println("["+topic+"] theta size: "+theta.size)
        val positive = doc._1.contains(topic)
        theta = updateTheta(doc._2, theta, lRate, positive)
    }
    println("["+topic+"] theta size: "+theta.size)

  }
  def classify(doc: XMLDocument) : Double = {
    val prob = probRelevant(Main.lrFeatures(doc),theta)
    //println("Probability: "+prob)
    prob
  }

  override def toString = theta.toList.toString()

  // HELPER FUNCTIONS

  def innerProduct(v1: Map[String,Double], v2: Map[String,Double]) : Double = {
    (v1.keySet & v2.keySet).map(k => (v1.getOrElse(k, 0.0) * v2.getOrElse(k, 0.0))).sum
  }

  def probRelevant(dFeature: Map[String,Double], theta: Map[String,Double]) : Double = {
    val result = 1.0 / (1.0 + Math.exp(-1.0 * bias - innerProduct(dFeature,theta)))
    //println(result)
    result
  }

  def scalarMultVector(scalar: Double, vector: Map[String,Double]) : Map[String,Double] = {
    vector.mapValues(i => i*scalar)
  }

  def vectorAdd(v1: Map[String,Double], v2: Map[String,Double]) : Map[String,Double] = {
    (v1.keySet ++ v2.keySet).map(k => ((k -> (v1.getOrElse(k, 0.0) + v2.getOrElse(k, 0.0)) )) ).toMap
  }

  def deltaTheta(dFeature: Map[String,Double], theta: Map[String,Double], rel: Boolean) : Map[String,Double] = {
    if(rel){
      scalarMultVector(1.0 - probRelevant(dFeature,theta), dFeature)
    }else{
      scalarMultVector(-1.0*probRelevant(dFeature,theta), dFeature)
    }
  }

  def updateTheta(doc: Map[String,Double], theta: Map[String,Double], lRate: Double, rel: Boolean) : Map[String,Double] = {
    vectorAdd(scalarMultVector(lRate, deltaTheta(doc,theta,rel)), theta)
  }

}