Model Tutorial

Fundamentally a Model is a container for Factors. Its primary function is, given a collection of variables, to return the Factors that neighbor those variables.

The trait Model leaves abstract how this mapping from Variables to Factors is maintained.

package cc.factorie.tutorial
object TutorialModel extends App {
  import cc.factorie._
  import cc.factorie.la._
  import cc.factorie.variable.BooleanVariable
  import cc.factorie.model.{ ItemizedModel, DotFactorWithStatistics2 }

Let’s start by creating some Variables and Factor classes. *

  val outputs: Seq[BooleanVariable] = for (i <- 0 until 10) yield new BooleanVariable
  val inputs: Seq[BooleanVariable] = for (i <- 0 until 10) yield new BooleanVariable(i % 2 == 0)
  val markovWeights = new DenseTensor2(Array(Array(1.0, 0.0), Array(0.0, 1.0)))
  class MarkovFactor(b1: BooleanVariable, b2: BooleanVariable) extends DotFactorWithStatistics2(b1, b2) {
    def weights = markovWeights
    override def factorName = "MarkovFactor"
  }
  val inputWeights = new DenseTensor2(Array(Array(1.0, -1.0), Array(-1.0, 1.0)))
  class InputFactor(bi: BooleanVariable, bo: BooleanVariable) extends DotFactorWithStatistics2(bi, bo) {
    def weights = inputWeights
    override def factorName = "InputFactor"
  }

  // ItemizedModel stores a given set of Factors, with their relations to Variables indexed by HashMaps.
  val m1 = new ItemizedModel
  m1 ++= (for (i <- 0 until 9) yield new MarkovFactor(outputs(i), outputs(i + 1)))
  m1 ++= inputs.zip(outputs).map({ case (i: BooleanVariable, o: BooleanVariable) => new InputFactor(i, o) })

  val f1 = m1.factors(outputs)
  println(f1)
}