Heatmap for "long" DataFrames

[metanoid]

The available heatmap plot from Plots.jl is built with a particular "shape" of data in mind: the user will define:

• x - the set of m unique x axis labels
• y - the set of n unique y axis labels
• z - an m-by-n shape of data values (usually a Matrix)

But this isn't the shape of the data in the most common scenarios in which I want a heatmap. The shape I usually find myself in is having a DataFrame with

• an X column with n copies of m x-axis labels
• a Y column with m copies of n y-axis labels
• a Z column of m-by-n data values, one for each distinct X and Y pair

In other words, the DataFrame is "long" rather than "wide".

What I think would be nice is some "recipe" in which the user can provide:

• x - a column to use for x-axis labels
• y - a column to use for y axis labels
• z - a column to use for colouring

Right now, my workaround is to use unstack on the DataFrame before plotting, but this is a bit ungainly.

[mkborregaard]

That makes sense, should be pretty easy to implement

[daschw]

That could/should be done for contour, surface and wireframe as well.

[poglesbyg]

Any update on this issue?