Please add support for filled-in contour plots

[jumutc]

I want to create in Gadfly something similar to imagesc/contourf in Winston or Matlab, but using Geom.rectbin seems to me incomplete and cannot to some full extent handle this functionality. Please find several inconsistencies provided in the plot (alike white regions). Please consider the following snippet of my code:

using MAT
vars = matread(joinpath("..", "data","ripley.mat"))
X = vars["X"]'
y = vars["Y"]

range1 = linspace(-1.5,1.5,200)
range2 = linspace(-0.5,1.5,200)
Xtest = meshgrid(range1,range2)'
diff1 = range1[2] - range1[1]
diff2 = range2[2] - range2[1]

yhat = ... #do predictions for points
yplot = ...#do predictions for plot 
yplot = yplot - minimum(yplot)
yplot = 2*(yplot ./ maximum(yplot)) - 1

using DataFrames
df = DataFrame()
df[:Xmin] = Xtest[1,:][:] 
df[:Ymin] = Xtest[2,:][:]
df[:Xmax] = Xtest[1,:][:] + diff1
df[:Ymax] = Xtest[2,:][:] + diff2
df[:class] = yplot[:]

using Gadfly
set_default_plot_size(20cm, 20cm)
plot(layer(x=X[1,yhat.>0], y=X[2,yhat.>0], Geom.point, Theme(default_color=color("orange"))),
     layer(x=X[1,yhat.<0], y=X[2,yhat.<0], Geom.point, Theme(default_color=color("black"))),
     layer(df, x_min="Xmin", x_max="Xmax", y_min="Ymin", y_max="Ymax", color="class", Geom.rectbin))

[jumutc]

plot(z=reshape(yplot,200,200), Geom.contour, xmin=[-1.5], xmax=[1.5], ymin=[-.5], ymax=[1.5]) is giving something else too

[darwindarak]

Geom.contour indexes the z matrix as [xcoord, ycoord]. So you should get the correct contour if you transpose z:

x = linspace(-1.5, 1.5, 200)
y = linspace(-0.5, 1.5, 200)
plot(z=reshape(yplot,200,200)', x=coords, y=coords, Geom.contour)

Do you mind trying (separately)

• remove the layers with points
• transpose your yplot
• shift x and y

to help narrow down the source of the problem?

[jumutc]

Hi, thanks for the suggestion.. By using this indexing I can get: But in reality I'm more interested in getting filled-in contour plot, like the one I could get with Geom.rectbin

[dcjones]

@jumutc Is the main issue you have with Geom.rectbin the white lines that you're seeing?

There's an SVG trick I used that aligns the rectangles to the nearest pixel, but it seems that can cause issues in a dense grid like in your plot, which explains the white lines. If that's fixed, would it otherwise be sufficient for your needs?

[jumutc]

Hi, @dcjones. That would be perfect for me. But there is also an issue with completely blank regions (outside of y_max, x_max etc) which sometimes appear on the figure. If the latter can be fixed quickly I would be very pleased. Thanks in advance!

[CorySimon]

+1, contour plots filled in with color would be very helpful. (probably a wrapper around rectbin).