Jitter reproducibility in geom_jitter()

future_changes.md

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

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

@@ -11,13 +11,20 @@ import org.jetbrains.letsPlot.core.plot.base.DataPointAesthetics
 import org.jetbrains.letsPlot.core.plot.base.GeomContext
 import org.jetbrains.letsPlot.core.plot.base.PositionAdjustment
 
-class JitterDodgePos(aesthetics: Aesthetics, groupCount: Int, width: Double?, jitterWidth: Double?, jitterHeight: Double?) :
+class JitterDodgePos(
+ aesthetics: Aesthetics,
+ groupCount: Int,
+ width: Double?,
+ jitterWidth: Double?,
+ jitterHeight: Double?,
+ jitterSeed: Long? = null
+) :
  PositionAdjustment {
  private val myJitterPosHelper: PositionAdjustment
  private val myDodgePosHelper: PositionAdjustment
 
  init {
- myJitterPosHelper = JitterPos(jitterWidth, jitterHeight)
+ myJitterPosHelper = JitterPos(jitterWidth, jitterHeight, jitterSeed)
  myDodgePosHelper = DodgePos(aesthetics, groupCount, width)
  }
 

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))
  }
 
@@ -34,7 +35,6 @@ internal class JitterPos(width: Double?, height: Double?) : PositionAdjustment {
  }
 
  companion object {
-
  val DEF_JITTER_WIDTH = 0.4
  val DEF_JITTER_HEIGHT = 0.4
  }

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 {
@@ -64,9 +64,10 @@ object PositionAdjustments {
  groupCount: Int,
  width: Double?,
  jitterWidth: Double?,
- jitterHeight: Double?
+ jitterHeight: Double?,
+ jitterSeed: Long?
  ): PositionAdjustment {
- return JitterDodgePos(aesthetics, groupCount, width, jitterWidth, jitterHeight)
+ return JitterDodgePos(aesthetics, groupCount, width, jitterWidth, jitterHeight, jitterSeed)
  }
 
  enum class Meta(private val handlesGroups: Boolean) {

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 {
@@ -115,12 +115,19 @@ abstract class PosProvider {
  }
  */
 
- fun jitterDodge(width: Double?, jitterWidth: Double?, jitterHeight: Double?): PosProvider {
+ fun jitterDodge(width: Double?, jitterWidth: Double?, jitterHeight: Double?, jitterSeed: Long?): PosProvider {
  return object : PosProvider() {
  override fun createPos(ctx: PosProviderContext): PositionAdjustment {
  val aesthetics = ctx.aesthetics
  val groupCount = ctx.groupCount
- return PositionAdjustments.jitterDodge(aesthetics, groupCount, width, jitterWidth, jitterHeight)
+ return PositionAdjustments.jitterDodge(
+ aesthetics,
+ groupCount,
+ width,
+ jitterWidth,
+ jitterHeight,
+ jitterSeed
+ )
  }
 
  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 {
@@ -478,6 +480,7 @@ object Option {
  const val DODGE_WIDTH = "dodge_width"
  const val JITTER_WIDTH = "jitter_width"
  const val JITTER_HEIGHT = "jitter_height"
+ const val JITTER_SEED = "jitter_seed"
  }
 
  object Stack {

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(
@@ -47,7 +48,8 @@ internal object PosProto {
  JITTER_DODGE -> PosProvider.jitterDodge(
  opts.getDouble(Pos.JitterDodge.DODGE_WIDTH),
  opts.getDouble(Pos.JitterDodge.JITTER_WIDTH),
- opts.getDouble(Pos.JitterDodge.JITTER_HEIGHT)
+ opts.getDouble(Pos.JitterDodge.JITTER_HEIGHT),
+ opts.getLong(Pos.JitterDodge.JITTER_SEED)
  )
 
  else -> throw IllegalArgumentException("Unknown position adjustments name: '$posName'")

python-package/lets_plot/plot/geom.py

@@ -4593,9 +4593,9 @@ def geom_density2df(mapping=None, *, data=None, stat=None, position=None, show_l
 
 
 def geom_jitter(mapping=None, *, data=None, stat=None, position=None, show_legend=None, sampling=None, tooltips=None,
- width=None,
- height=None,
+ width=None, height=None,
  color_by=None, fill_by=None,
+ seed=None,
  **other_args):
  """
  Display jittered points, especially for discrete plots or dense plots.
@@ -4637,6 +4637,9 @@ def geom_jitter(mapping=None, *, data=None, stat=None, position=None, show_legen
  Define the color aesthetic for the geometry.
  fill_by : {'fill', 'color', 'paint_a', 'paint_b', 'paint_c'}, default='fill'
  Define the fill aesthetic for the geometry.
+ seed : int
+ A random seed to make the jitter reproducible.
+ If None (the default value), the seed is initialised with a random value.
  other_args
  Other arguments passed on to the layer.
  These are often aesthetics settings used to set an aesthetic to a fixed value,
@@ -4697,6 +4700,7 @@ def geom_jitter(mapping=None, *, data=None, stat=None, position=None, show_legen
  ggplot({'x': x, 'y': y}, aes(x='x', y='y')) + \\
  geom_jitter(aes(color='x', size='y'), \\
  sampling=sampling_random(n=600, seed=60), \\
+ seed=37, + \\
  show_legend=False, width=.25) + \\
  scale_color_grey(start=.75, end=0) + \\
  scale_size(range=[1, 3])
@@ -4712,6 +4716,7 @@ def geom_jitter(mapping=None, *, data=None, stat=None, position=None, show_legen
  tooltips=tooltips,
  width=width, height=height,
  color_by=color_by, fill_by=fill_by,
+ seed=seed,
  **other_args)
 
 

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):
@@ -185,7 +188,7 @@ def position_nudge(x=None, y=None):
  return _pos('nudge', x=x, y=y)
 
 
-def position_jitterdodge(dodge_width=None, jitter_width=None, jitter_height=None):
+def position_jitterdodge(dodge_width=None, jitter_width=None, jitter_height=None, jitter_seed=None):
  """
  This is primarily used for aligning points generated through `geom_point()`
  with dodged boxplots (e.g., a `geom_boxplot()` with a fill aesthetic supplied).
@@ -204,6 +207,9 @@ def position_jitterdodge(dodge_width=None, jitter_width=None, jitter_height=None
  Jittering height.
  The value of `jitter_height` is relative and typically ranges between 0 and 0.5.
  Values that are greater than 0.5 lead to overlapping of the points.
+ jitter_seed : int
+ A random seed to make the jitter reproducible.
+ If None (the default value), the seed is initialised with a random value.
 
  Returns
  -------
@@ -232,10 +238,11 @@ def position_jitterdodge(dodge_width=None, jitter_width=None, jitter_height=None
  stat='boxplot') + \\
  geom_point(aes(x='c', y='x', color='c'), \\
  size=4, shape=21, fill='white',
- position=position_jitterdodge())
+ position=position_jitterdodge(jitter_seed=42))
 
  """
- return _pos('jitterdodge', dodge_width=dodge_width, jitter_width=jitter_width, jitter_height=jitter_height)
+ return _pos('jitterdodge', dodge_width=dodge_width, jitter_width=jitter_width, jitter_height=jitter_height,
+ jitter_seed=jitter_seed)
 
 
 def position_stack(vjust=None, mode=None):