Gadfly.jl plots

To plot the Chains via Gadfly.jl, use the DataFrames constructor:

using DataFrames
using CategoricalArrays
using Gadfly
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)

plot(df, x=:A, color=:chain, Geom.density, Guide.ylabel("Density"))

Multiple parameters

Or, to show multiple parameters in one plot, use DataFrames.stack

sdf = stack(df, names(chn), variable_name=:parameter)
first(sdf, 5)

5 rows × 4 columns

and Gadfly.Geom.subplot_grid

plot(sdf, ygroup=:parameter, x=:value, color=:chain,
    Geom.subplot_grid(Geom.density), Guide.ylabel("Density"))

This is very flexible. For example, we can look at the first two chains only by using DataFrames.filter

first_chain = filter([:chain] => c -> c == 1 || c == 2, sdf)

plot(first_chain, xgroup=:parameter, ygroup=:chain, x=:value,
    Geom.subplot_grid(Geom.density, Guide.xlabel(orientation=:horizontal)),
    Guide.xlabel("Parameter"), Guide.ylabel("Chain"))

Trace

plot(first_chain, ygroup=:parameter, x=:iteration, y=:value, color=:chain,
    Geom.subplot_grid(Geom.point), Guide.ylabel("Sample value"))