Learn.oms

val images = Val[Array[File]]
val imageDirectory = Val[File]
val nbTrees   = Val[Int]
val treeDepth = Val[Int]
val kFold = 10

val random = new util.Random(42)
val imagesArrays: Array[Array[File]] = 
  (0 until kFold).map(i => random.shuffle((workDirectory / "images").listFiles).toArray[File]).toArray

val learningOutput = Val[String]

val prepareFiles = 
  ScalaTask("""
  val imageDirectory = newDir("image")
  for 
    (f, i) <- images.zipWithIndex
  do f.copy(imageDirectory / s"image$i.jpg")
  """) set (
    (inputs, outputs) += (nbTrees, treeDepth),
    inputs += images,
    outputs += imageDirectory
  )

val learning = PythonTask(
  workDirectory / "forest.py",
  arguments = "images ${nbTrees} ${treeDepth}",
  stdOut = learningOutput,
  libraries = Seq("matplotlib", "scikit-learn", "pillow", "scikit-image")
) set (
  inputs += (nbTrees, treeDepth),
  inputFiles += (imageDirectory, "images", true),
  outputs += (nbTrees, treeDepth)
)

DirectSampling(
  sampling =   
    (nbTrees in (5 to 25 by 5)) x
    (treeDepth in (3 to 18 by 3)),
  evaluation = 
    DirectSampling(
      sampling = (images in imagesArrays),
      evaluation = prepareFiles -- learning 
    ) hook (workDirectory / "points.csv", Seq(nbTrees, treeDepth, learningOutput))
)