Merge b992189 into 974d3f5

docs/Project.toml

@@ -1,4 +1,5 @@
 [deps]
+CairoMakie = "13f3f980-e62b-5c42-98c6-ff1f3baf88f0"
 CategoricalArrays = "324d7699-5711-5eae-9e2f-1d82baa6b597"
 DataFrames = "a93c6f00-e57d-5684-b7b6-d8193f3e46c0"
 Documenter = "e30172f5-a6a5-5a46-863b-614d45cd2de4"
@@ -9,6 +10,7 @@ MLJDecisionTreeInterface = "c6f25543-311c-4c74-83dc-3ea6d1015661"
 StatsPlots = "f3b207a7-027a-5e70-b257-86293d7955fd"
 
 [compat]
+CairoMakie = "0.6"
 CategoricalArrays = "0.8, 0.9, 0.10"
 DataFrames = "0.22, 1"
 Documenter = "0.26, 0.27"

docs/make.jl

@@ -15,6 +15,7 @@ makedocs(
         "Getting started" => "getting-started.md",
         "Plotting" => [
             "StatsPlots.jl" => "statsplots.md",
+            "Makie.jl" => "makie.md",
             "Gadfly.jl" => "gadfly.md"
         ],
         "API" => [

docs/src/gadfly.md

@@ -9,15 +9,6 @@ using Gadfly
 write_svg(path, p; w=6inch, h=4inch) = Gadfly.draw(Gadfly.SVG(path, w, h), p) # hide
 using MCMCChains
 
-# Define the experiment.
-n_iter = 100
-n_name = 3
-n_chain = 2
-
-# Experiment results.
-val = randn(n_iter, n_name, n_chain) .+ [1, 2, 3]'
-val = hcat(val, rand(1:2, n_iter, 1, n_chain))
-
 chn = Chains(randn(100, 2, 3), [:A, :B])
 df = DataFrame(chn)
 df[!, :chain] = categorical(df.chain)

docs/src/makie.md

@@ -0,0 +1,66 @@
+# Makie.jl plots
+
+Compared to Gadfly.jl and StatsPlots.jl, Makie.jl is the most flexible plotting library that you can use.
+On this page is an example function for plotting with Makie.jl which you can use directly or further tweak to your needs.
+
+```@example makie
+using CairoMakie
+using DataFrames
+using MCMCChains
+
+chns = Chains(randn(300, 5, 3), [:A, :B, :C, :D, :E])
+```
+
+```@example makie
+function plot_chains(chns; density_func=density!)
+    params = names(chns, :parameters)
+
+    df = DataFrame(chns)
+    n_chains = length(unique(df.chain))
+    n_samples = nrow(df) / n_chains
+
+    # Alternatively, use `CategoricalArrays.categorical`.
+    df[!, :chain] = string.(df.chain)
+
+    fig = Figure(; resolution=(1_000, 800))
+
+    # Create and store separate axes for showing iterations.
+    values_axs = [Axis(fig[i, 1]; ylabel=string(c)) for (i, c) in enumerate(params)]
+    for (ax, col) in zip(values_axs, params)
+        for i in 1:n_chains
+            chain = string(i)
+            values = filter(:chain => ==(chain), df)[:, col]
+            lines!(ax, 1:n_samples, values; label=chain)
+        end
+    end
+
+    # Thanks to having stored the axes before, we can apply some extra tweaks.
+    # These tweaks usually depend on the kind of model being fitted.
+    values_axs[end].xlabel = "Iteration"
+    hideydecorations!.(values_axs; label=false)
+    hidexdecorations!.(values_axs[1:end-1]; grid=false)
+
+    # Create and store separate axes for showing the parameter density estimate.
+    density_axs = [Axis(fig[i, 2]; ylabel=string(c)) for (i, c) in enumerate(params)]
+    for (ax, col) in zip(density_axs, params)
+        for i in 1:n_chains
+            chain = string(i)
+            values = filter(:chain => ==(chain), df)[:, col]
+            density_func(ax, values; label=chain)
+        end
+    end
+
+    # Just like above, we add some extra tweaks.
+    density_axs[end].xlabel = "Parameter estimate"
+    linkxaxes!(density_axs...)
+    hideydecorations!.(density_axs)
+    hidexdecorations!.(density_axs[1:end-1]; grid=false)
+
+    return fig
+end
+nothing # hide
+```
+
+```@example makie
+plot_chains(chns)
+```