Add vertical bands like vspan() in Plots.jl

[mm3509]

I would like to easily plot vertical bands, e.g. as a background for a recession in a time-series plot. Plots.jl has vspan(). In Gadfly, it requires a lot of code and mixes the color of the bands with the color of the time-series line.

For reference, the code is below, copied from StackOverflow:

using Gadfly, DataFrames, Colors

df = DataFrame(year = [1990; 2000], x = [0; 1], color = [1; 1])

x_shade = [1995 1995 1996 1996]
y_shade = [0 1 1 0]

theme = Theme(
    discrete_highlight_color = u -> ARGB(1, 1, 1, 0),
    default_color = colorant"grey")

p = plot(
    layer(
        x = x_shade,
        y = y_shade,
        Geom.polygon(preserve_order = true, fill = true),
        order = 1
    ),
    layer(df,
          y = "x",
          x = "year",
          color = "color",
          Geom.line,
          order = 2
          ),
    theme
)

[bjarthur]

this produces identical output:

plot(df, y = "x", x = "year", color = "color", Geom.line,
     xintercept=[1995.5], Geom.vline(color="grey",size=[10]))

the only problem is that size can not currently be specified in the plot coordinate system.

[mm3509]

Thanks @bjarthur ! In my specific case, that it is a problem: I am judging the behavior of a time-series depending around the edges of those boxes. So I would need to specify it the size in the original plot, e.g. size = 1. I would rather avoid hacks to convert between the plot coordinate system and the aesthetics system.

[Mattriks]

Geom.rect can also be used:

using RDatasets, Gadfly

Dp = dataset("ggplot2","presidential")[3:end,:]
De = dataset("ggplot2","economics")
De[:Unemploy] /= 10^3
Dp[:y1] = 0; Dp[:y2] = 12
theme(x) = style(default_color=parse(Colors.Colorant, x))

p = plot(Dp, xmin=:Start, xmax=:End, ymin=:y1, ymax=:y2, Geom.rect, color=:Party,
    layer(De, x=:Date, y=:Unemploy, Geom.line, order=2, theme("black")),
    Scale.color_discrete_manual("deepskyblue", "lightcoral"),
    Coord.cartesian(xmin=Date("1965-01-01"), ymax=12),
    Guide.xlabel("Time"), Guide.ylabel("Unemployment (x10³)"), Guide.colorkey(title=""),
    Theme(key_position=:top)
)

[Mattriks]

And here is similar plot code in R, for comparison:

p = ggplot(presidential)+geom_rect(aes(xmin=start, xmax=end, ymin=-Inf, ymax=Inf, fill=party))+
    geom_line(data=economics, aes(x=date, y=unemploy))+
    scale_fill_manual(values=c("deepskyblue", "lightcoral"))+
    coord_cartesian(xlim=as.Date(c("1965-01-01", "2009-12-31")), ylim=c(0,12000), expand=F)+
    labs(x="Time", y="Unemployment", fill="")+
    theme(legend.position="top")

[Mattriks]

And another way, which uses Geom.vline (as in @bjarthur's example), but with context units: cx, cy from Compose. Yes, this allows size to be specified in the same units as x, y. In the plot below, note how the bar width matches the point distance = 1 year.

using Compose, DataFrames, Gadfly
srand(123)
df = DataFrame(year=1990:2000, x = rand(11))

p = plot(df, x=:year, y=:x, xintercept=[1993.5, 1995.5, 1997.5], 
    Geom.line, Geom.point, Geom.vline(color="lightgrey", size=[1cx]))

Should add this to the docs!

[bjarthur]

of course! i should've thought of that!! @miguelmorin if this solves your problem please close this issue

[mm3509]

Thanks @Mattriks ! Your solution with Geom.rect worked for me.