spacings in hstack/vstack/gridstack

[IgorDouven]

Is there a way to fine-tune spacings when using hstack/vstack/gridstack? I tried to control this via plot_padding for the individual plots that are being stacked, but this doesn't allow me to reduce vertical space between the rows (which is badly needed in my plot). Something like Spacings in Mathematica's GraphicsGrid would be nice to have.

[Mattriks]

Here is a wip vstack Mk 2 function - feedback welcome!

using Colors, Compose, DataFrames, Gadfly

function vstack2(plots::Vector{Plot}; spacing::Float64=0.0, heights::Vector{<:Real}=Float64[])
    n = length(plots)
    heights==Float64[] && (heights = fill(1/n, n))
    sumh = cumsum([0;heights])
    vpos = sumh + spacing.*[0; 1:n-1; n-1]
    M = [(context(0,v,1,h), render(p)) for (v,h,p) in zip(vpos[1:n], heights, plots)]
    return compose(context(units=UnitBox(0,0,1,vpos[n+1])), M...)
end

srand(123)
D = vcat([DataFrame(x=1:5, y=rand(5), g=g) for g in [" A "," B "," C "," D "]]...)

bgcol = RGBA(0.99,0.96,0.9,0.4)
pa = plot(D[1:10,:], y=:y, xgroup=:g, Geom.subplot_grid(layer(x=:x, Geom.point, Geom.hair)), style(panel_fill=bgcol))
pb = plot(D[11:20,:], y=:y, xgroup=:g, Geom.subplot_grid(layer(x=:x, Geom.point, Geom.hair)))

# spacing and heights are in the same "relative" units
p = vstack2([pa,pb,pa], spacing=-0.8, heights=[3,4,3])
draw(PNG(6inch,6inch),p)

[IgorDouven]

Thanks, but this didn't work for me. I would now have to first use hstack to collect some plots into rows and then I would want to use vstack2 to collate those rows. However, in that case the input for the latter function is of the wrong type (Compose.Context instead of Vector{Plot}). Is there a way around this?

[Mattriks]

I tested this one with hstacked plots:

using Compose, Gadfly

function vstack2(plots::Vector{T}; spacing::Float64=0.0, heights::Vector{<:Real}=Float64[]) where T<:Union{Context,Plot}
    n = length(plots)
    heights==Float64[] && (heights = ones(n))
    sumh = cumsum([0;heights])
    vpos = sumh + spacing.*[0; 1:n-1; n-1]
    i = isa.(plots, Plot)
    any(i) && (plots[i] = render.(plots[i]))
    M = [(context(0,v,1,h), p)  for (v,h,p) in zip(vpos[1:n], heights, plots)]
    return compose(context(units=UnitBox(0,0,1,vpos[n+1])), M...)
end

[IgorDouven]

This worked for me as well -- great many thanks! I'm closing the issue now, though it would be nice to have this function in the documentation.

[bjarthur]

let's leave this open as adding this functionality would be useful

[Mattriks]

Yes I would like to open a PR for this sometime (with new functions for Gadfly.vstack, Gadfly.hstack and Gadfly.gridstack). There is an issue here that currently Gadfly.{hv}stack use Compose.{hv}stack, but the new functions (e.g. vstack2 above) do not use Compose.{hv}stack. I think Compose.{hv}stack should remain as they are¹, The question then is: what is the best way of incorporating the new Gadfly._stack functions into Gadfly? e.g. should they be renamed?

¹Just adding that I think the idea here is to discontinue the use of the Compose._stack functions in Gadfly, but leave the Compose._stack functions as they are, for people who are using Compose to make graphics/pictures.

[bjarthur]

i'd suggest making a new function, say figure(), which when given a column vector of Plots acts like vstack, given a row vector like hstack, and given a matrix like gridstack. keyword arguments could replace title and specify spacing.

would be great to also be able to do insets, as well as to align plots by their axis origins.

see https://cran.r-project.org/web/packages/egg/vignettes/Ecosystem.html for how R implements all of this functionality.

my only hesitation is that this approach is not declarative.

[Mattriks]

I like the suggestion of a figure() function (tbd). I view it as a helper function, for users who want that. If you want declarative graphics than that functionality is in Compose e.g. here is an inset plot:

using Compose, DataFrames, Gadfly
pa = plot(XPI, x=:Year, y=:TCC, Geom.line,
    Coord.cartesian(ymax=150),
    Theme(default_color="red", panel_fill="CornSilk") )
pb = plot(XPI, x=:Year, y=:CPI, Geom.line, Coord.cartesian(ymin=10),
    Guide.ylabel("CPI (\$US '000s)"),
    Guide.title("Xmas Price Index
(https://en.wikipedia.org/wiki/Christmas_Price_Index)"),
    Guide.annotation(compose(context(0w,0h,0.75w,0.5h), render(pa))),
    Theme(default_color="green")
)

[jonathanBieler]

One issue with figure is that it's used to open and manage figures in Matlab-like figure manager. But since it's either without argument or just an index it could be fine.

What about a inset function for inset, that work like layer with additional position/size arguments ?

plot(
    layer(x=x, y=y, Geom.point)
    inset(x=z, y=w, Geom.line, position=(0w,0h,0.75w,0.5h) ),
)

[Mattriks]

In Gadfly Grammar of Graphics, my plot above could be done perhaps like this:

plot( layer(x=Year, y=CPI, Geom.line, Theme(...)),
    layer(x=Year, y=TCC, Geom.inset(Geom.line, pos=(0.0w,0.0h,0.75w,0.5h)), Theme(...) ) 
)

similar in syntax to Geom.subplot_grid().

[CNOT]

Let me put this here too. It would be really nice, if one had the option to use the same legend or title or axis labels for a stack of plots. Geom.subplot_grid assumes there is some sort of categorized structure to your data, so it doesn't satisfy the simple cases where you basically just want to display the same data in different scales without repeating the legend.

Also, I don't know how to use the compose workarounds for a legend.

After some googling, I figured an R's ggplot2 approach would be to extract the legend from one plot and then create a custom layout and place each plot and the legend in their corresponding cells. grid_arrange_shared_legend

Is there a way to do the same in Julia's Gadfly?

[Mattriks]

Here's my quick attempt at such a plot:

using Compose, DataFrames, Gadfly, RDatasets
mtcars = sort(dataset("datasets", "mtcars"), :Cyl)

myplot(D) = plot(D, x=:MPG, y=:WT, color=:Cyl, Geom.point, 
    Scale.color_discrete, Theme(key_position=:none, plot_padding=[1mm]))
Dp = by(mtcars, :Cyl, myplot) 

p1 = plot(mtcars, x=:MPG, y=:WT, color=:Cyl, Geom.point, 
    Scale.color_discrete, Theme(key_position=:none, plot_padding=[1mm]))

p2 = gridstack(reshape(Union{Plot,Compose.Context}[Dp.x1; context()], 2, 2))

gck = Guide.manual_color_key("", ["4","6","8"].*"\t\t", Scale.color_discrete().f(3))
p3 = compose(context(0.4,0.5), render(gck, Theme(), Gadfly.Aesthetics())[1].ctxs[3])

# vstack2 is in a post above
vstack2([hstack(p1, p2); p3], heights=[9,1])

There are some color issues in p2, and the development of a Geom.subplot_wrap would help (#855).

[Mattriks]

Example 2

gck = Guide.manual_color_key("Cylinders", ["4","6","8"].*"\t\t", Scale.color_discrete().f(3))
p3 = compose(context(),(context(0.5,0), render(gck, Theme(), Gadfly.Aesthetics())[1].ctxs[1]))
p2 = gridstack(reshape(Union{Plot,Compose.Context}[Dp.x1; p3], 2, 2))
hstack(p1, p2)

[CNOT]

This actually helps a lot. Thanks.