Jitter reproducibility in geom_jitter()

future_changes.md

@@ -5,4 +5,4 @@
 ### Changed
 
 ### Fixed
-
+- Jitter reproducibility in geom_jitter [[#911](https://github.com/JetBrains/lets-plot/issues/911)].

plot-base/src/commonMain/kotlin/org/jetbrains/letsPlot/core/plot/base/pos/JitterPos.kt

@@ -12,20 +12,21 @@ import org.jetbrains.letsPlot.core.plot.base.GeomContext
 import org.jetbrains.letsPlot.core.plot.base.PositionAdjustment
 import kotlin.random.Random
 
-internal class JitterPos(width: Double?, height: Double?) : PositionAdjustment {
+internal class JitterPos(width: Double?, height: Double?, seed: Long? = null) : PositionAdjustment {
 
  //uniform distribution
  private val myWidth: Double
  private val myHeight: Double
+ private val random: Random = seed?.let { Random(seed) } ?: Random.Default
 
  init {
  myWidth = width ?: DEF_JITTER_WIDTH
  myHeight = height ?: DEF_JITTER_HEIGHT
  }
 
  override fun translate(v: DoubleVector, p: DataPointAesthetics, ctx: GeomContext): DoubleVector {
- val x = (2 * Random.nextDouble() - 1) * myWidth * ctx.getResolution(Aes.X)
- val y = (2 * Random.nextDouble() - 1) * myHeight * ctx.getResolution(Aes.Y)
+ val x = (2 * random.nextDouble() - 1) * myWidth * ctx.getResolution(Aes.X)
+ val y = (2 * random.nextDouble() - 1) * myHeight * ctx.getResolution(Aes.Y)
  return v.add(DoubleVector(x, y))
  }
 

plot-base/src/commonMain/kotlin/org/jetbrains/letsPlot/core/plot/base/pos/PositionAdjustments.kt

@@ -51,8 +51,8 @@ object PositionAdjustments {
  return FillPos(aesthetics, vjust, stackingMode)
  }
 
- fun jitter(width: Double?, height: Double?): PositionAdjustment {
- return JitterPos(width, height)
+ fun jitter(width: Double?, height: Double?, seed: Long?): PositionAdjustment {
+ return JitterPos(width, height, seed)
  }
 
  fun nudge(width: Double?, height: Double?): PositionAdjustment {

plot-builder/src/commonMain/kotlin/org/jetbrains/letsPlot/core/plot/builder/assemble/PosProvider.kt

@@ -85,10 +85,10 @@ abstract class PosProvider {
  }
  }
 
- fun jitter(width: Double?, height: Double?): PosProvider {
+ fun jitter(width: Double?, height: Double?, seed: Long?): PosProvider {
  return object : PosProvider() {
  override fun createPos(ctx: PosProviderContext): PositionAdjustment {
- return PositionAdjustments.jitter(width, height)
+ return PositionAdjustments.jitter(width, height, seed)
  }
 
  override fun handlesGroups(): Boolean {

plot-stem/src/commonMain/kotlin/org/jetbrains/letsPlot/core/spec/GeomProto.kt

@@ -113,6 +113,7 @@ class GeomProto(val geomKind: GeomKind) {
  Meta.NAME to PosProto.JITTER,
  Pos.Jitter.WIDTH to layerOptions.getDouble(Geom.Jitter.WIDTH),
  Pos.Jitter.HEIGHT to layerOptions.getDouble(Geom.Jitter.HEIGHT),
+ Pos.Jitter.SEED to layerOptions.getLong(Geom.Jitter.SEED)
  )
 
  Y_DOT_PLOT -> if (layerOptions.hasOwn(Geom.YDotplot.STACKGROUPS) &&

plot-stem/src/commonMain/kotlin/org/jetbrains/letsPlot/core/spec/Option.kt

@@ -247,6 +247,7 @@ object Option {
  object Jitter {
  const val WIDTH = "width"
  const val HEIGHT = "height"
+ const val SEED = "seed"
  }
 
  object Step {
@@ -467,6 +468,7 @@ object Option {
  object Jitter {
  const val WIDTH = "width"
  const val HEIGHT = "height"
+ const val SEED = "seed"
  }
 
  object Nudge {

plot-stem/src/commonMain/kotlin/org/jetbrains/letsPlot/core/spec/PosProto.kt

@@ -36,7 +36,8 @@ internal object PosProto {
  FILL -> configureFillPosition(opts)
  JITTER -> PosProvider.jitter(
  opts.getDouble(Pos.Jitter.WIDTH),
- opts.getDouble(Pos.Jitter.HEIGHT)
+ opts.getDouble(Pos.Jitter.HEIGHT),
+ opts.getLong(Pos.Jitter.SEED)
  )
 
  NUDGE -> PosProvider.nudge(

python-package/lets_plot/plot/pos.py

@@ -96,7 +96,7 @@ def position_dodgev(height=None):
  return _pos('dodgev', height=height)
 
 
-def position_jitter(width=None, height=None):
+def position_jitter(width=None, height=None, seed=None):
  """
  Adjust position by assigning random noise to points. Better for discrete values.
 
@@ -110,6 +110,9 @@ def position_jitter(width=None, height=None):
  Jittering height.
  The value of height is relative and typically ranges between 0 and 0.5.
  Values that are greater than 0.5 lead to overlapping of the points.
+ seed : int
+ A random seed to make the jitter reproducible.
+ If None (the default value), the seed is initialised with a random value.
 
  Returns
  -------
@@ -137,10 +140,10 @@ def position_jitter(width=None, height=None):
  ggplot({'x': x, 'y': y, 'c': c}, aes('x', 'y')) + \\
  geom_point(aes(fill='c'), show_legend=False, \\
  size=8, alpha=.5, shape=21, color='black', \\
- position=position_jitter(width=.2, height=.2))
+ position=position_jitter(width=.2, height=.2, seed=42))
 
  """
- return _pos('jitter', width=width, height=height)
+ return _pos('jitter', width=width, height=height, seed=seed)
 
 
 def position_nudge(x=None, y=None):