StatPlots groupedbar order x axis

stelmo · August 23, 2018, 12:46am

Hi,

Is there a way to order the x-axis bars using groupedbar? Currently it is placing the individual bars alphabetically, but I would like to specify the order manually. I have tried using plot!(xticks=(1:10, labels)), but that only changes the actual labels and doesn’t move the bars…

mkborregaard · August 23, 2018, 8:36am

No, and you can’t specify the order of the individual sub-bars either. But you should be able to do that IMHO. If you can think of a good set of keywords and interface to specify this (perhaps group could take a tuple where the second arg is the group order, and there could be an xorder argument? not sure) you should open an issue on StatPlots and we’ll implement this. Perhaps @piever has a good idea, he’s thought a lot about these things.

piever · August 23, 2018, 9:53am

I though categorical arrays used to work for this? Normally to specify the order of your data you can put your array into a CategoricalArray (see here). I think in that case Plots should sort by the custom order.

mkborregaard · August 23, 2018, 3:00pm

That works for the groups, but unfortunately not for the x argument. I guess the solution is to implement that in StatPlots and add it to the readme.

piever · August 23, 2018, 3:50pm

Yes that needs fixing: if the user specifies an order with the CategoricalArray we should respect that in the x axis: am a bit confused as to why it doesn’t work already, but will investigate.

stelmo · August 23, 2018, 3:57pm

I see you added the issue, thanks!

mkborregaard · August 23, 2018, 6:58pm

Yes, perhaps this needs fixing in Plots rather than StatPlots? Even if it means taking on a dep on CategoricalArrays (but that shouldn’t be necessary, no?)

piever · August 23, 2018, 10:21pm

I also think this should be fixed on Plots side but it shouldn’t need extra dependency: we should instruct Plots to respect the order of the custom type before converting to string / float

piever · August 24, 2018, 10:20am

I’m actually no longer sure I completely understand what you are trying to achieve, could you post the code you’re using to generate the plot?

I’m a bit confused because I don’t think the order used by default for the x axis is alphabetical so I wanted to make sure I understand the issue.

Related to this, we could probably allow to use groupedbar(["a", "a", "b", "b"], [1,2,3,4]) to give the same plot as groupedbar(["a", "b"], [1 2; 3 4]), currently it errors.

mkborregaard · August 24, 2018, 1:30pm

Agreed.

stelmo · August 24, 2018, 5:50pm

Sure, so in this example I would like to have the x-axis arranged B, C, A (as an arbitrary example). By using xtick=(…,…) it changes only the actual labels, it doesn’t change the columns. So the column should move with the label.

using Plots
using StatPlots
using GroupedErrors
using DataFrames
gr()

df = DataFrame([[“A”,“B”,“A”,“B”,“A”,“B”,“C”,“C”,“C”], [1,2,3,2,3,4,6,6,8]], [:Observation, :Count])
@> df begin
@across :all
@x _.Observation :discrete

@y _.Count
@plot groupedbar(ylabel="Fraction of proteome", legend=true)

end

Does this make sense?

piever · August 25, 2018, 5:38pm

Minor remark: use the </> button to format your code, the " button makes a quotation and it looks a bit funky.

Interestingly enough GroupedErrors already does the correct thing and you can use CategoricalArrays:


using Plots
using StatPlots
using GroupedErrors
using DataFrames
using CategoricalArrays
gr()
v = CategoricalArray(["A","B","A","B","A","B","C","C","C"])
levels!(v, ["B", "C", "A"])
df = DataFrame([v, [1,2,3,2,3,4,6,6,8]], [:Observation, :Count])
@> df begin
    @across :all
    @x _.Observation :discrete

    @y _.Count
    @plot groupedbar(ylabel="Fraction of proteome", legend=true)
end

See the CategoricalArrays docs for more details. I think you should also be able to make a column categorical from the DataFrame directly, something like:

df = DataFrame([["A","B","A","B","A","B","C","C","C"], [1,2,3,2,3,4,6,6,8]], [:Observation, :Count])
categorical!(df, :Observation)
levels!(df[:Observation], ["B", "C", "A"])

piever · August 25, 2018, 5:47pm

Solving the thing in general for StatPlots is a bit trickier: should groupedbar(x, y, group=z) return the x axis in the order it is encountered or should it sort it, like boxplot and violinplot do? Maybe we could sort it.

I almost think plotting commands where sorting is possible should take the same keyword arguments as sort (rev, by) to allow reversing the order or some custom sorting.

stelmo · August 25, 2018, 8:39pm

Okay, this fix works for GR, but when using PyPlot it doesn’t show the actual x-axis labels (see the image, although I can manually fix this using the xticks attribute). For both Plotly and PlotlyJS it just crashes. Thanks for all your help so far!

piever · August 25, 2018, 9:06pm

This sounds like a bug on Plots side, could you open an issue over there?

stelmo · August 25, 2018, 9:16pm

Will do!

newb_gk · October 22, 2019, 5:59pm

Is there an update on this? I’m having the same sorting issue of the x axis with grouped bars on StatsPlots. It doesn’t seem to be an issue on Plots itself, but couldn’t figure out how to plot grouped bars with Plots using any of the backends.

JonasIsensee · June 12, 2020, 9:03am

Hi,

so I just figured out how to do it after a lot of digging.

The issue is that in
RecipesPipeline.jl there is
group_labels = sort(collect(unique(v)))

github.com

JuliaPlots/RecipesPipeline.jl/blob/304db22c33b64a902dd91606a8886ec371f62809/src/group.jl#L11


      
          # # Grouping
          
          "A special type that will break up incoming data into groups, and allow for easier creation of grouped plots"
          mutable struct GroupBy
              group_labels::Vector                # length == numGroups
              group_indices::Vector{Vector{Int}}  # list of indices for each group
          end
          
          # this is when given a vector-type of values to group by
          function _extract_group_attributes(v::AVec, args...; legend_entry = string)
              group_labels = sort(collect(unique(v)))
              n = length(group_labels)
              if n > 100
                  @warn("You created n=$n groups... Is that intended?")
              end
              group_indices = Vector{Int}[filter(i -> v[i] == glab, eachindex(v)) for glab in group_labels]
              GroupBy(map(legend_entry, group_labels), group_indices)
          end
          
          legend_entry_from_tuple(ns::Tuple) = join(ns, ' ')

Now in theory there are CategoricalArrays for this purpose,
however unique(ctg::CategoricalArray) returns an ordinary Vector{String} instead of a new CategoricalArray.

Defining this solves the problem in this case

function Base.unique(ctg::CategoricalArray)
    l = levels(ctg)
    newctg = CategoricalArray(l)
    levels!(newctg, l)
end

ctg = CategoricalArray(["a", "b", "c", "a", "b", "c"])
levels!(ctg, ["c", "b", "a"])
groupedbar(1:6, group=ctg)

test

jlawrie · August 30, 2022, 10:00pm

so I just figured out how to do it after a lot of digging.

Thank you!

@JonasIsensee Just so as you know, in response to this similar question on GitHub for StatsPlots.jl, I quickly wrote up your method into a function which adapts the usual inputs to StatsPlots.groupedbar, meaning that the ordering of the names and categories is preserved.

I credited you directly but I wasn’t able to tag you on GitHub, so I thought it would be kind to let you know.

Have a nice day!