Performance of skipmissings

testperf.jl

using BenchmarkTools, Missings

function make_x_y(N, p)
    x = map(1:N) do _
      rand() < p ? missing : rand()
    end
    y = map(1:N) do _
        rand() < p ? missing : rand()
    end
    sx = skipmissing(x)
    mx, my = skipmissings(x, y)
    cx = [xx === missing ? rand() : xx for xx in x]
    cy = [yy === missing ? rand() : yy for yy in y]
    return (x = x, y = y, sx = sx, mx = mx, my = my, cx = cx, cy = cy)
end

function testperfzip(X, Y)
    r = zero(eltype(X)) * zero(eltype(Y))
    for (x, y) in zip(X, Y)
        r += x * y
    end
    r
end

function testperfzip_manual(X, Y)
    r = zero(Missings.nonmissingtype(eltype(X))) * zero(Missings.nonmissingtype(eltype(Y)))
    for (x, y) in zip(X, Y)
        if x !== missing && y !== missing
            r += x * y
        end
    end
    r
end

function manual_sum_view(x, y)
  sum(view(x, .!ismissing.(x) .& .!ismissing.(y)))
end

function manual_sum_index(x, y)
  sum(x[.!ismissing.(x) .& .!ismissing.(y)])
end

p = 0

t_100 = make_x_y(100, p)
t_10_000 = make_x_y(10_000, p)

macro custombtime(expr)
  :(println("    ", @sprintf("%.10e", @belapsed $expr), " seconds\n"))
end

println("============")
println("Proportion missing: $p")
println("============\n")

println("Sum:")
println("==========")
println("N = 100: skipmissings")
@custombtime sum($t_100.mx)
println("N = 100: vector")
@custombtime sum($t_100.cx)
println("N = 100: manual sum with view")
@custombtime manual_sum_view($t_100.x, $t_100.y)
println("N = 100: manual sum with index")
@custombtime manual_sum_index($t_100.x, $t_100.y)


println("N = 10_000: skipmissings")
@custombtime sum($t_10_000.mx)
println("N = 10_000: vector")
@custombtime sum($t_10_000.cx)
println("N = 10_000: manual sum with view")
@custombtime manual_sum_view($t_10_000.x, $t_10_000.y)
println("N = 10_000: manual sum with index")
@custombtime manual_sum_index($t_10_000.x, $t_10_000.y)


println("")
println("Zip")
println("==========")
println("N = 100: skipmissings")
@custombtime testperfzip($t_100.mx, $t_100.my)
println("N = 100: vector")
@custombtime testperfzip($t_100.cx, $t_100.cy)
println("N = 100: manual skipping missings")
@custombtime testperfzip($t_100.x, $t_100.y)

println("N = 10_000: skipmissings")
@custombtime testperfzip($t_10_000.mx, $t_10_000.my)
println("N = 10_000, vector")
@custombtime testperfzip($t_10_000.cx, $t_10_000.cy)
println("N = 10_000: manual skipping missings")
@custombtime testperfzip($t_10_000.x, $t_10_000.y)


println("")