Heatmaps (clustering), dendrograms/trees

[versipellis]

I'd love to see heatmaps support Clustering.jl's Hierarchical Clustering, sort of something like R's own heatmap.2 (https://www.rdocumentation.org/packages/gplots/versions/3.0.1.1/topics/heatmap.2) where the clustering can be plotted as dendrograms in addition to the heatmaps. I imagine something like this could be done with layouts and line primitives, but I'm only just getting started on Makie and am not sure if that's possible.

[asinghvi17]

I think there's a Plots recipe for it, so it should be possible using Makie; however, the layouting will be a bit of a pain until we can implement a better system.

[KronosTheLate]

Any news? Would love to have makie recipes for dendrograms and trees.

[SimonDanisch]

Someone needs to write a recipe for it... I don't know of anyone working on it right now!

[jbrea]

I was also looking for this, and found an easy solution to plot dendrograms (without a recipe) using a modified version of StatsPlots.treeposition:

function treepositions(hc::Hclust; useheight = true, orientation = :vertical)
    order = StatsBase.indexmap(hc.order)
    nodepos = Dict(-i => (float(order[i]), 0.0) for i in hc.order)
    xs = []
    ys = []
    for i in 1:size(hc.merges, 1)
        x1, y1 = nodepos[hc.merges[i, 1]]
        x2, y2 = nodepos[hc.merges[i, 2]]
        xpos = (x1 + x2) / 2
        ypos = useheight ?  hc.heights[i] : (max(y1, y2) + 1)
        nodepos[i] = (xpos, ypos)
        push!(xs, [x1, x1, x2, x2])
        push!(ys, [y1, ypos, ypos, y2])
    end
    if orientation == :horizontal
        return ys, xs
    else
        return xs, ys
    end
end
function dendrogram(h; color = :blue, kwargs...)
    CairoMakie.Axis(Figure()[1, 1])
    for (x, y) in zip(treepositions(h; kwargs...)...)
        lines!(x, y; color)
    end
    current_figure()
end

[pascalschulthess]

Very nice functions. Thanks @jbrea. How would I implement colouring the lines in the dendrogram by cluster membership after I defined the number of clusters and determined the membership with e.g. cutree(h, k=4)?

[jbrea]

Good question @pascalschulthess :) I guess I would modify the treepositions function to return not only a list of xs and ys but also a list of colors. Based on the output of cutree and the hc structure it should be possible to determine the correct color for each node. But it would take me some time to figure everything out.

[kescobo]

See here for relevant discussion attached to StatsPlots regarding adding colors.

I'd like to get this recipe built - have you put any additional effort into making a PR for this, or a separate package? I'm happy to take over, but don't want to butt in if you have it in hand.

[jbrea]

I didn't work on this, but it would be nice to have it @kescobo !