hclustrecipe.jl

function hclustplot(hc::Hclust, useheight::Bool)
    o = indexmap(hc.order)
    n = [x for x in 1:length(o)]

    pos = treepositions(hc, useheight)


    xs = []
    ys = []
    for i in 1: size(hc.merge, 1)
        x1 = pos[hc.merge[i,1]][1]
        x2 = pos[hc.merge[i,2]][1]
        append!(xs, [x1,x1,x2,x2])

        y1 = pos[hc.merge[i,1]][2]
        y2 = pos[hc.merge[i,2]][2]
        useheight ? h = hc.height[i] : h = 1
        newy = maximum([y1,y2]) + h
        append!(ys, [y1,newy,newy,y2])
    end
    return (reshape(xs, 4, size(hc.merge, 1)), reshape(ys, 4, size(hc.merge, 1)))
end

function treepositions(hc::Hclust, useheight::Bool)
    order = indexmap(hc.order)
    positions = Dict{}()
    for (k,v) in order
        positions[-k] = (v, 0)
    end
    for i in 1:size(hc.merge,1)
        xpos = mean([positions[hc.merge[i,1]][1], positions[hc.merge[i,2]][1]])
        if hc.merge[i,1] < 0 && hc.merge[i,2] < 0
            useheight ? ypos = hc.height[i] : ypos = 1
        else
            useheight ? h = hc.height[i] : h = 1
            ypos = maximum([positions[hc.merge[i,1]][2], positions[hc.merge[i,2]][2]]) + h
        end

        positions[i] = (xpos, ypos)
    end
    return positions
end

I found that the indexmap was in DataFrames package. it is easy to make a customized one though.

indexmap and mean are now in StatsBase and Statistics packages respectively. So simply add using StatsBase, Statistics to the start of the recipe. Or simply using StatsKit.