Fix Array conversion of ChainDataFrame (#188)

* Drop column with parameter names with undefined values in Array conversion

* Allow to specify lags as AbstractVector

* Always forward sections keyword argument to summarize

* Test plotting with UnicodePlots

* Update Project.toml

.travis.yml

@@ -16,10 +16,6 @@ julia:
   - 1.3
   - nightly
 
-# deactivate plot display
-env:
-  - GKSwstype=nul
-
 matrix:
   allow_failures:
     - julia: nightly

Project.toml

@@ -3,7 +3,7 @@ uuid = "c7f686f2-ff18-58e9-bc7b-31028e88f75d"
 keywords = ["markov chain monte carlo", "probablistic programming"]
 license = "MIT"
 desc = " Chain types and utility functions for MCMC simulations."
-version = "3.0.2"
+version = "3.0.3"
 
 [deps]
 AbstractMCMC = "80f14c24-f653-4e6a-9b94-39d6b0f70001"
@@ -33,8 +33,10 @@ julia = "^1"
 [extras]
 DataFrames = "a93c6f00-e57d-5684-b7b6-d8193f3e46c0"
 KernelDensity = "5ab0869b-81aa-558d-bb23-cbf5423bbe9b"
+Logging = "56ddb016-857b-54e1-b83d-db4d58db5568"
 StatsPlots = "f3b207a7-027a-5e70-b257-86293d7955fd"
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
+UnicodePlots = "b8865327-cd53-5732-bb35-84acbb429228"
 
 [targets]
-test = ["KernelDensity", "StatsPlots", "Test", "DataFrames"]
+test = ["DataFrames", "KernelDensity", "Logging", "StatsPlots", "Test", "UnicodePlots"]

src/plot.jl

@@ -58,8 +58,8 @@ const supportedplots = push!(collect(keys(translationdict)), :mixeddensity, :cor
     end
 
     if st == :autocorplot
-        lags = 0:(maxlag === nothing ? round(Int, 10 * log10(length(range(c)))) : maxlag)
-        ac = autocor(c, lags=collect(lags); showall=true)
+        lags = 0:(maxlag === nothing ? round(Int, 10 * log10(length(range(c)))) : maxlag)        
+        ac = autocor(c, lags=lags; showall=true)
         ac_mat = convert(Array, ac)
         val = colordim == :parameter ? ac_mat[:, :, i]' : ac_mat[i, :, :]
         _AutocorPlot(lags, val)

src/stats.jl

@@ -17,7 +17,7 @@ The `digits` keyword may be a(n)
 - `Dict`, with a similar structure as `NamedTuple`. `Dict(mean => 2, std => 3)` would set `mean` to two digits and `std` to three digits.
 """
 function autocor(chn::Chains;
-        lags::Vector=[1, 5, 10, 50],
+        lags::AbstractVector{<:Integer}=[1, 5, 10, 50],
         demean::Bool=true,
         relative::Bool=true,
         showall=false,
@@ -33,6 +33,7 @@ function autocor(chn::Chains;
     return summarize(chn, funs...;
         func_names = func_names,
         showall = showall,
+        sections = sections,
         append_chains = append_chains,
         name = "Autocorrelation",
         digits=digits)
@@ -200,6 +201,7 @@ function hpd(chn::Chains; alpha::Real=0.05,
     return summarize(chn, u, l;
         func_names = labels,
         showall=showall,
+        sections=sections,
         name="HPD",
         digits=digits)
 end
@@ -235,9 +237,9 @@ function quantile(chn::Chains;
     end
 
     return summarize(chn, funs...;
-        sections=sections,
         func_names=func_names,
         showall=showall,
+        sections=sections,
         name="Quantiles",
         digits=digits,
         append_chains=append_chains, 
@@ -285,7 +287,7 @@ function ess(chn::Chains;
     # Misc allocations.
     m = n_chain_orig * 2
     maxlag = min(maxlag, n-1)
-    lags = collect(0:maxlag)
+    lags = 0:maxlag
 
     # Preallocate B, W, varhat, and Rhat vectors for each param.
     B = Vector(undef, length(param))
@@ -411,9 +413,9 @@ function summarystats(chn::Chains;
 
     # Summarize.
     summary_df = summarize(chn, funs...;
-        sections=sections,
         func_names=func_names,
         showall=showall,
+        sections=sections,
         name="Summary Statistics",
         additional_df = ess_df,
         digits=digits,
@@ -450,9 +452,9 @@ function mean(chn::Chains;
 
     # Summarize.
     summary_df = summarize(chn, funs...;
-        sections=sections,
         func_names=func_names,
         showall=showall,
+        sections=sections,
         name="Mean",
         digits=digits)
 

src/summarize.jl

@@ -163,21 +163,23 @@ function Base.lastindex(c::ChainDataFrame, i::Integer)
 end
 
 function Base.convert(::Type{Array}, c::C) where C<:ChainDataFrame
-    arr = Array{Float64, 2}(undef, c.n_rows, c.n_cols)
-    ks = collect(keys(c.nt))
+    T = promote_eltype_namedtuple_tail(c.nt)
+    arr = Array{T, 2}(undef, c.n_rows, c.n_cols - 1)
 
-    for i in 2:c.n_cols
-        arr[:, i] = c.nt[ks[i]]
+    for (i, k) in enumerate(Iterators.drop(keys(c.nt), 1))
+        arr[:, i] = c.nt[k]
     end
 
     return arr
 end
 
 Base.convert(::Type{Array{ChainDataFrame,1}}, cs::Array{ChainDataFrame,1}) = cs
 function Base.convert(::Type{Array}, cs::Array{C,1}) where C<:ChainDataFrame
-    arrs = [convert(Array, cs[j]) for j in 1:length(cs)]
-    return cat(arrs..., dims = 3)
+    return mapreduce((x, y) -> cat(x, y; dims = Val(3)), cs) do c
+        reshape(convert(Array, c), Val(3))
+    end
 end
+
 """
 
 # Summarize a Chains object formatted as a DataFrame

src/utils.jl

@@ -122,6 +122,18 @@ Base.@pure function merge_union_types(names::Tuple{Vararg{Symbol}}, a::Type{<:Na
     return Tuple{types...}
 end
 
+# promote element types of the tail of a NamedTuple
+function promote_eltype_namedtuple_tail(::NamedTuple{k,v}) where {k,v}
+    return promote_eltype_tuple_type(Base.tuple_type_tail(v))
+end
+
+# promote element types of a tuple
+promote_eltype_tuple_type(::Type{Tuple{}}) = Any
+promote_eltype_tuple_type(::Type{Tuple{T}}) where T = T
+function promote_eltype_tuple_type(t::Type{<:Tuple})
+    Base.promote_eltype(Base.tuple_type_head(t), promote_eltype_tuple_type(Base.tuple_type_tail(t)))
+end
+
 """
     cskip(x)
 

test/plot_test.jl

@@ -2,6 +2,10 @@ using Test
 using StatsPlots
 using MCMCChains
 
+import Logging
+
+unicodeplots()
+
 n_iter = 500
 n_name = 3
 n_chain = 2
@@ -11,60 +15,51 @@ val = hcat(val, rand(1:2, n_iter, 1, n_chain))
 
 chn = Chains(val)
 
-@testset "Plotting tests" begin
+# Silence all warnings.
+level = Logging.min_enabled_level(Logging.current_logger())
+Logging.disable_logging(Logging.Warn)
 
+@testset "Plotting tests" begin
     # plotting singe plotting types
     println("traceplot")
-    ps_trace = traceplot(chn, 1)
-    @test isa(ps_trace, Plots.Plot)
-
+    display(traceplot(chn, 1))
+    println()
+    
     println("meanplot")
-    ps_mean = meanplot(chn, 1)
-    @test isa(ps_mean, Plots.Plot)
-
+    display(meanplot(chn, 1))
+    println()
+    
     println("density")
-    ps_density = density(chn, 1)
-    @test isa(ps_density, Plots.Plot)
-
-    ps_density = density(chn, 1, append_chains=true)
-    @test isa(ps_density, Plots.Plot)
-
+    display(density(chn, 1))
+    display(density(chn, 1, append_chains=true))
+    println()
+
     println("autocorplot")
-    ps_autocor = autocorplot(chn, 1)
-    @test isa(ps_autocor, Plots.Plot)
-
+    display(autocorplot(chn, 1))
+    println()
+    
     #ps_contour = plot(chn, :contour)
 
     println("histogram")
-    ps_hist = histogram(chn, 1)
-    @test isa(ps_hist, Plots.Plot)
-
-    println("mixeddensity")
-    ps_mixed = mixeddensity(chn, 1)
-    @test isa(ps_mixed, Plots.Plot)
-
+    display(histogram(chn, 1))
+    println()
+
+    println("\nmixeddensity")
+    display(mixeddensity(chn, 1))
+
     # plotting combinations
-    ps_trace_mean = plot(chn)
-    @test isa(ps_trace_mean, Plots.Plot)
-
-    ps_trace_mean = plot(chn, append_chains=true)
-    @test isa(ps_trace_mean, Plots.Plot)
-
-    savefig("demo-plot.png")
-
-    ps_mixed_auto = plot(chn, seriestype = (:mixeddensity, :autocorplot))
-    @test isa(ps_mixed_auto, Plots.Plot)
+    display(plot(chn))
+    display(plot(chn, append_chains=true))
+    display(plot(chn, seriestype = (:mixeddensity, :autocorplot)))
 
     # Test plotting using colordim keyword
-    p_colordim = plot(chn, colordim = :parameter)
-    @test isa(p_colordim, Plots.Plot)
-
+    display(plot(chn, colordim = :parameter))
+
     # Test if plotting a sub-set work.s
-    p_subset = plot(chn, 2)
-    @test isa(p_subset, Plots.Plot)
-
-    p_subset_colordim = plot(chn, 2, colordim = :parameter)
-    @test isa(p_subset_colordim, Plots.Plot)
-
-    rm("demo-plot.png")
+    display(plot(chn, 2))
+    display(plot(chn, 2, colordim = :parameter))
+    println()
 end
+
+# Reset log level.
+Logging.disable_logging(level)