Test Utilities


Thorough testing is inevitable to assure high quality when implementing your own algorithms on top of Graph for Scala. The package scalax.collection.generator, part of the module graph-core, helps you to achieve this goal by providing facilities for user-friendly random graph generation.

Besides accepting a node generator, Graph for Scala Test allows you to control the metrics of random graphs including their order, density and density distribution. Further, you control which edge types should connect nodes including custom weight and label generators.

The companion objects also provide some predefined generators such as smallConnectedIntDi.

Creating random graphs

Think of RandomGraph as the native implementation to generate any kind of Graph with random nodes and edges satisfying user-defined graph metrics.

Obtaining generators for graphs with predefined metrics

The simplest way to obtain a random graph is to use a predefined generator like

import scalax.collection.generator._

val predefined = RandomGraph.tinyConnectedIntDi(Graph)
val tinyGraph = predefined.draw // Graph[Int,DiEdge]

Setting individual graph metrics

You are certainly not stick with the metrics applied to predefined generators but you can adjust them individually:

object sparse_1000_Int extends RandomGraph.IntFactory {
  val order = 1000
  val nodeDegrees = NodeDegreeRange(1,10)
  override def connected = false
val randomSparse = RandomGraph[Int,UnDiEdge,Graph](
    Graph, sparse_1000_Int, Set(UnDiEdge))
val sparseGraph = randomSparse.draw // Graph[Int,UnDiEdge]

As to

  1. Defines the metrics for graphs of Int nodes.
  2. The generated graph will have an order of 1000.
  3. Node degrees will be within the range of 1 and 10.
  4. Allows the generated graph to be disconnected.
  5. A RandomGraph generator is instantiated to generate graphs with the above metrics and the edge type UnDiEdge.
  6. Draws a concrete random graph from the generator.

Obtaining generators for individual graph types

As you have noticed the previous generator was restricted to the node type of Int. But to cover your use case you usually need to obtain random graphs with a specific node type. Let's define a class Person(name: String, yearOfBirth: Int) for this purpose. Also, let's create PersonData containing meaningful data to Persons:

object PersonData {
  val firstNames = Set(
      "Alen", "Alice", "Bob", "Jack", "Jim",
      "Joe", "Kate", "Leo", "Tim", "Tom").to[Vector]
  val firstNamesSize = firstNames.size
  val surnames = Set(
      "Bell", "Brown", "Clark", "Cox", "King",
      "Lee", "Moore", "Ross", "Smith", "Wong").to[Vector]
  val surnamesSize = surnames.size
  def order = firstNamesSize * surnamesSize / 10
  def degrees = new NodeDegreeRange(2, order - 2)
  val maxYearOfBirth = 2010
import scalax.collection.edge.LDiEdge
case class Person(name: String, yearOfBirth: Int)
object Person {
  import PersonData._
  private val r = new scala.util.Random
  def drawFirstName: String = firstNames(r.nextInt(firstNamesSize))  
  def drawSurame: String = surnames(r.nextInt(surnamesSize))

  def drawName: String = s"$drawFirstName, $drawSurame"
  def drawYearOfBirth = maxYearOfBirth - r.nextInt(100)
val randomMixedGraph =
    Graph, new RandomGraph.Metrics[Person] {
      val order = PersonData.order
      val nodeDegrees = PersonData.degrees
      def nodeGen: Person = Person(Person.drawName, Person.drawYearOfBirth)
    Set(UnDiEdge, LDiEdge)
val mixedGraph = randomMixedGraph.draw

mixedGraph will now have undirected and labeled directed edges looking like

  Person(Alice, Smith,1967),
  Person(Kate, Ross,1921),
  Person(Leo, Bell,2008),
  Person(Leo, Smith,1983),
  Person(Alice, Smith,1967)~>Person(Kate, Ross,1921) 'C,
  Person(Leo, Bell,2008)~Person(Leo, Smith,1983),

You can even supply your own label and weight generators.

Keeping metric constraints in mind

You may wonder why we divided the number of all possible combinations of first names and surnames by 10. The reason is that RandomGraph must be able to draw distinct nodes. Ok, they become most probably distinct due to the presence of yearOfBirth. But had we just Person(name: String) and def order = firstNamesSize * surnamesSize, we'd get more and more duplicate names when drawing them. Therefore RandomGraph is implemented to cope with duplicates up to a certain limit. Above that limit the generator assumes that the supplied node generator behaves ill and rejects to generate the graph by throwing an exception. Otherwise it would take the risk of an infinite loop.

In conclusion, always plan your node generator with respect to the graph order carefully. By rule of thumb, allow your node generator to have about ten-fold capacity over the graph order.

Node degrees are another important design aspect. Generating graphs with a high density may fail.

Integrating with ScalaCheck

GraphGen, a thin wrapper around RandomGraph, is provided for out-of-the-box ScalaCheck integration. Below you'll find the ScalaCheck equivalent of the previous examples. We assume some knowledge of ScalaCheck.

Obtaining Arbitrary instances for graphs with predefined metrics

type IntDiGraph = Graph[Int,DiEdge]
implicit val arbitraryTinyGraph = GraphGen.tinyConnectedIntDi[Graph](Graph)
val properTiny = forAll(arbitrary[IntDiGraph]) { g: IntDiGraph =>
    g.order == GraphGen.TinyInt.order

Setting individual graph metrics for Arbitrary instances

object Sparse_1000_Int extends GraphGen.Metrics[Int] {
  val order = 1000
  val nodeDegrees = NodeDegreeRange(1,10)
  def nodeGen: Gen[Int] = Gen.choose(0, 10 * order)
  override def connected = false
type IntUnDiGraph = Graph[Int,UnDiEdge]
implicit val arbitrarySparseGraph = Arbitrary {
  GraphGen[Int,UnDiEdge,Graph](Graph, Sparse_1000_Int, Set(UnDiEdge)).apply
val properSparse = forAll(arbitrary[IntUnDiGraph]) { g: IntUnDiGraph =>
  g.order == Sparse_1000_Int.order

Obtaining Arbitrary instances to generate individual graph types

import scalax.collection.edge.LDiEdge
case class Person(name: String, yearOfBirth: Int)
object Person {
  import PersonData._
  def firstNameGen : Gen[String] = Gen.oneOf(firstNames)  
  def surameGen: Gen[String] = Gen.oneOf(surnames)

  def nameGen: Gen[String] = Gen.resultOf( (firstName: String, surname: String) =>
    s"$firstName, $surname")(
    Arbitrary(firstNameGen), Arbitrary(surameGen))

  def yearOfBirthGen: Gen[Int] = Gen.choose(maxYearOfBirth - 100, maxYearOfBirth)
object MixedMetrics extends GraphGen.Metrics[Person] {
  val order = PersonData.order
  val nodeDegrees = PersonData.degrees
  def nodeGen: Gen[Person] = Gen.resultOf( (name: String, year: Int) =>  
    Person(name, year))(
    Arbitrary(Person.nameGen), Arbitrary(Person.yearOfBirthGen))

type Mixed = Graph[Person,UnDiEdge]
implicit val arbitraryMixedGraph = Arbitrary {
      Graph, MixedMetrics, Set(UnDiEdge, LDiEdge)).apply

val properMixedGraph = forAll(arbitrary[Mixed]) { g: Mixed =>
  g.order == MixedMetrics.order

Limiting the minimum number of successful test

As you know, by default ScalaCheck executes 100 tests before it takes your property proved. You may desire to limit the number of tests because, for instance, you are dealing with bigger graphs. Here is an example for how you can achieve this in the context of ScalaTest:

import org.scalatest.{Matchers, Spec}
import org.scalatest.prop.PropertyChecks
class TGraphGenTest
    extends Spec with Matchers with PropertyChecks {
  implicit val config =
    PropertyCheckConfig(minSuccessful = 5, maxDiscarded = 5)
  object `generated Tiny graph` { 
    implicit val arbitraryTinyGraph =
    def `should conform to tiny metrics` {
      forAll(arbitrary[IntDiGraph]) { g: IntDiGraph =>
        g.order should equal (GraphGen.TinyInt.order)