Respect CategoricalArray order in `group` kwarg

[kescobo]

l = CategoricalArray(repeat(["a", "c", "b"], 3))
levels!(l, ["a", "c", "b"])
c = [:red :blue :green]

scatter(rand(9), rand(9),
    group=l, # I want "c" to go second
    color=c) # I want "c" to be blue

Instead, "c" should be second, and be blue:

scatter(rand(9), rand(9), primary=false, color=repeat(vec(c), 3))
for i in eachindex(levels(l))
    scatter!([],[], label=levels(l)[i], color=c[i])
end

[mkborregaard]

Agreed, that should be the behaviour.

[kescobo]

Looking back at this - sorry to be so terse! There was a discussion in slack that led to it. Don't want to seem like I'm just coming in here and being demanding 😄

[mkborregaard]

Nonono, you're being a good open source citizen by reporting it and I do agree :-)

[pgbarletta]

I was about to open an issue about this same topic. In my case, I'm having trouble with groupedbar() so I can't plot() and then plot!(). Unless there is some other way of bypassing this. Is there?

[kescobo]

@pgbarletta Can you explain a bit more about what you're trying to do with grouped bar? I'm not sure, but since that's part of StatPlots (and has some nuances since it's a user recipe rather than a series recipe), it might be worth opening a separate issue over there with a MWE

[pgbarletta]

I actually made an example before finding this issue. Hope it's clear enough:

https://nbviewer.jupyter.org/github/pgbarletta/lbp/blob/master/src/Untitled.ipynb

C1-CG and O1, O2 are atom names, so my molecule is getting reordered :)

Let me know if I should open an issue on StatPlots. Thanks for the help.

[kescobo]

Thanks for the example @pgbarletta. I think your case is a bit different - this issue is about CategoricalArrays, and specifically about the order of the groups, (in your case V1 and V2) rather than the order on the xaxis.

In your case, you can just use numbers for ordering and then set the labels:

groupedbar(repeat(1:18, 2), hcat(v1, v2),
    group = g_lig, xaxis=("Atoms", (1:18, x_lig)))

[pgbarletta]

You're right, it worked! I didn't know xaxis could take a Tuple and assign labels to the ticks.
Thanks a lot.

https://nbviewer.jupyter.org/github/pgbarletta/lbp/blob/master/src/Untitled.ipynb

[WMOkiishi]

It seems that this now does just work. However, there's one way in which it still doesn't quite do what I'd like it to.

l = CategoricalArray(repeat(["a", "c", "b"], 3))
levels!(l, ["a", "c", "d", "b"])  # Note the addition of an unused "d" value here
c = [:red :blue :orange :green]

scatter(rand(9), rand(9),
    group=l,  # Order should be a, c, (d,) b -- d is skipped, but the order is otherwise respected 
    color=c)  # Whether any "d" values are present or not, I'd like "b" to be green -- it's orange instead

I haven't been able to find an elegant workaround. The best I've come up with is using recipes with my own custom keywords for grouping and colors. As far as I can tell, since grouping is handled before any custom recipes I can add, there's no way for me to hook into that process.

[kescobo]

Hmm, blast from the past here... I'm mostly using Makie these days. Do you think it makes sense to close this issue and open a new one for your issue?

[WMOkiishi]

Yes, that makes sense to me.