Skip to main content

Configuration

Walker​

A walker is a simple scala case class that contains an id, a step count, and data :

case class Walker[T](
  id: Long,
  step: Long,
  data: T
)

But you can't make one by yourself; they are automatically generated by aruku thanks to the walker configuration.

Walker Configuration​

The walker configuration defines how many walkers will be doing random walks, how many epochs, how a walker is initialized, how it's updated, and how it's affected to a vertice.

modules/aruku/WalkerConfig.scala
case class WalkerConfig[T](
  numWalkers: Long,
  numEpochs: Int,
  parallelism: Int,
  init: VertexId => T,
  update: (Walker[T], VertexId, Edge[Double]) => T,
  start: StartingStrategy
)

It can be constant, as the walker is not updated at every step. DeepWalk is one example of such a walker configuration :

modules/aruku/WalkerConfig.scala
object WalkerConfig {

  def constant[T](
    numWalkers: Long,
    numEpochs: Int,
    parallelism: Int,
    init: VertexId => T,
    start: StartingStrategy
  )

}

Or dynamic if we need to update the data of the walker at every step.

modules/aruku/WalkerConfig.scala
object WalkerConfig {

  def updating[T](
    numWalkers: Long,
    numEpochs: Int,
    parallelism: Int,
    init: VertexId => T,
    update: (Walker[T], VertexId, Edge[Double]) => T,
    start: StartingStrategy
  ) 

}

Walker Engine Configuration​

You can configure the engine itself at the moment with the "spark.graphx.pregel.checkpointInterval" to break the lineage of the random walk if you are doing a lot of steps.