How to force DistributionChart to plot empty datasets?

Question

Is there any way to force DistributionChart to display anything where there is an empty dataset? In the below example I'd like to have 0-s appearing where data is missing, but even a simple Point would be sufficient as that can be labeled. The problem is that ChartElementFunction is not applied to empty datasets and thus Labeled wrappers are also ignored. Furthermore, ChartLabels (or association keys) cannot be forced to use different sub-labels for successive sets (it's always s1, s2, s3). I'd rather avoid reconstructing coordinates to use with Epilog (I've tried and failed with more complex BarSpacing values) or extract tick coordinates from InputForm[plot] or reconstruct built-in ChartElementFunctions (see ChartElementData@DistributionChart), as these methods are not very robust. data = <| "D1" -> <|"s1" -> {1, 2, 3, 4}, "s2" -> Labeled[{}, "Empty", Center], "s3" -> {3, 4, 5, 6}|>, "D2" -> <|"s1" -> {}, "s2" -> {}, "s3" -> {}|>, "D3" -> <|"s1" -> {3, 4}, "s2" -> {1, 2, 3, 4, 5, 6}, "s3" -> {}|>, "D4" -> <|"s1" -> {1, 2}, "s2" -> {3, 4}, "s3" -> {5, 6}|> |>; plot = DistributionChart[Map[Labeled[#, Length@#, Below] &, data, {2}], BarSpacing -> {.2, 1}, ChartLabels -> Automatic, ChartElementFunction -> "Density" ]

kglr · Accepted Answer

Update: a custom ChartElementFunction that can be combined with built-in ChartElementFunctions: ceF2[cedf_: ChartElementDataFunction["Density"]][vpos_: {0, 0}] := If[#3 == {"empty"}, {PointSize[Large], Point[Mean@Transpose@{#[[1]],vpos}]}, cedf[##]] & Examples: using a dataset that does not require version 10 functions SeedRandom[1] datac = {{RandomVariate[PoissonDistribution[5], 30], Labeled[{}, "Empty", Center], RandomVariate[NormalDistribution[], 50]}, {{}, {}, {0, 0}}, {RandomVariate[PoissonDistribution[3], 50], RandomChoice[{1, 2, 3, 4, 5, 6}, 100], {}}, {{1, 2}, {3, 4}, {5, 6}}}; vp0 = Min[datac /. Labeled | Style -> (#1 &)]; DistributionChart[Map[Labeled[# /. {} | Labeled[{}, __] :> ((vp = {vp0 - 1, vp0 - 1}) -> "empty"), Length[# /. Labeled -> (# &)], Below] &, datac, {2}], BaseStyle -> EdgeForm[], BarSpacing -> {.2, 1}, ChartLabels -> {"s1", "s2", "s3"}, ChartStyle -> 63, ChartElementFunction -> ceF2[][vp], ImageSize -> 600] Use ChartElementFunction -> ceF2[ChartElementDataFunction["GlassQuantile", "Quantile" -> 10, "QuantileShading" -> True]][vp] to get Original post: Your data modified to have both {} and {0,0} elements as well as elements with zero mean: datab= <|"D1" -> <|"s1" -> {1, 2, 3, 4}, "s2" -> Labeled[{}, "Empty", Center], "s3" -> {-3,-2,2,3}|>, "D2" -> <|"s1" -> {}, "s2" -> {}, "s3" -> {0,0}|>, "D3" -> <|"s1" -> {3, 4}, "s2" -> {1, 2, 3, 4, 5, 6}, "s3" -> {}|>, "D4" -> <|"s1" -> {1, 2}, "s2" -> {3, 4}, "s3" -> {5, 6}|> |>; Use metadata to distinguish empty sets, and modify the ChartElementFunction to render empty sets as Points: vp0 = Min[datab /. Labeled|Style->(#&)]; DistributionChart[Map[Labeled[#/.{}|Labeled[{},__]:> ((vp={vp0-1,vp0-1})->"empty"), Length[# /. Labeled ->(#&)], Below] &, datab, {2}], BarSpacing -> {.2, 1},ChartLabels ->{"s1", "s2" , "s3" }, ChartElementFunction -> (If[#3 == {"empty"}, {PointSize[Large],Point[Mean@Transpose@{#[[1]],vp}]}, Rectangle @@ Transpose@#1] &) ] Or define a custom ChartElementDataFunction ceF[vpos_:{0,0}]:=If[#3 == {"empty"}, PointSize[Large], Point[Mean@Transpose@{#[[1]],vpos}]}, Rectangle@@Transpose@#1] & and use as DistributionChart[Map[Labeled[#/.{}|Labeled[{},__]:> ((vp={vp0-1,vp0-1})->"empty"), Length[#/. Labeled[{},__]:>{}], Below] &, datab, {2}], BarSpacing -> {.2, 1}, ChartLabels ->{"s1", "s2" , "s3" }, ChartElementFunction -> ceF[vp] ] (* same picture *)

m_goldberg · Answer

Would this work for you? I need to assume the data can be transformed into

xData = 
   <|"D1" -> <|"s1" -> {1, 2, 3, 4}, "s2" -> {0, 0}, "s3" -> {3, 4, 5, 6}|>, 
     "D2" -> <|"s1" -> {0, 0}, "s2" -> {0, 0}, "s3" -> {0, 0}|>, 
     "D3" -> <|"s1" -> {3, 4}, "s2" -> {1, 2, 3, 4, 5, 6}, "s3" -> {0, 0}|>, 
     "D4" -> <|"s1" -> {1, 2}, "s2" -> {3, 4}, "s3" -> {5, 6}|>|>

Then the a distribution chart that indicates zero data items where the inner associations have value {0, 0}, can be drawn with

plot = DistributionChart[
  Map[Labeled[#, If[# == {0, 0}, 0, Length @ #], Below] &, xData, {2}],
  BarSpacing -> {.2, 1},
  ChartLabels -> Automatic,
  ChartElementFunction -> ((Rectangle @@ Transpose @ #) &)]

chart

Note: the original data can transformed in the form I use here with

xdata = data /. _Labeled -> {} /. {} -> {0, 0};

How to force DistributionChart to plot empty datasets?

2 Answers

Add your own answers!

Ask a Question