Improvements to ecdfplot

Project.toml

@@ -1,6 +1,6 @@
 name = "StatsPlots"
 uuid = "f3b207a7-027a-5e70-b257-86293d7955fd"
-version = "0.14.23"
+version = "0.14.24"
 
 [deps]
 Clustering = "aaaa29a8-35af-508c-8bc3-b662a17a0fe5"

src/ecdf.jl

@@ -2,20 +2,44 @@
 # ---------------------------------------------------------------------------
 # empirical CDF
 
-@recipe function f(ecdf::StatsBase.ECDF)
-    seriestype := :steppost
+function allsortedunique(x)
+    return all(eachindex(x)[1:end-1]) do i
+        @inbounds x[i] < x[i + 1]
+    end
+end
+
+@recipe function f(ecdf::StatsBase.ECDF; npoints=nothing)
+    seriestype --> :steppost
     legend --> :topleft
-    x = [ecdf.sorted_values[1]; ecdf.sorted_values]
-    if :weights in propertynames(ecdf) && !isempty(ecdf.weights)
-         # support StatsBase versions >v0.32.0
-        y = [0; cumsum(ecdf.weights) ./ sum(ecdf.weights)]
+    if npoints !== nothing
+        x = [-Inf; range(extrema(ecdf)...; length=npoints)]
     else
-        y = range(0, 1; length = length(x))
+        xnonunique = ecdf.sorted_values
+        xunique = allsortedunique(xnonunique) ? xnonunique : unique(xnonunique)
+        x = [-Inf; xunique]
     end
+    y = ecdf(x)
     x, y
 end
 
+
+"""
+    ecdfplot(x; npoints = nothing)
+
+Plot the empirical cumulative distribution function (ECDF) of `x`.
+
+By default, the ECDF is evaluated at the unique points in `x`. If `npoints` is provided,
+it is instead evaluated on a uniform grid of length `npoints` between the extrema of `x`.
+This is useful when `x` is large.
+
+```julia-repl
+julia> ecdfplot(randn(100))
+
+julia> ecdfplot(randn(1_000_000); npoints=100, seriestype=:path)
+```
+"""
 @userplot ECDFPlot
+
 recipetype(::Val{:ecdfplot}, args...) = ECDFPlot(args)
 @recipe function f(p::ECDFPlot)
     x = p.args[1]