fix warnings; fixes #90

src/CPDs/static_cpd.jl

@@ -25,7 +25,7 @@ function Distributions.fit(::Type{StaticCPD{D}},
  parents::NodeNames=NodeName[],
  ) where {D<:Distribution}
 
- d = fit(D, data[target])
+ d = fit(D, data[!,target])
  StaticCPD(target, parents, d)
 end
 

src/DiscreteBayesNet/tables.jl

@@ -14,9 +14,10 @@ end
 
 Base.convert(::Type{DataFrame}, t::Table) = t.potential
 
-for f in [:names, :unique, :size, :eltype, :setindex!, :getindex]
+for f in [:names, :unique, :size, :eltype, :setindex!]
  @eval (Base.$f)(t::Table, x...) = $f(t.potential, x...)
 end
+Base.getindex(t::Table, x...) = getindex(t.potential, !, x...)
 
 for s in [:(==), :(!=)]
  @eval (Base.$s)(t::Table, f::DataFrame) = $s(t.potential, f)

src/Factors/factors_dataframes.jl

@@ -9,7 +9,7 @@ Convert a Factor to a DataFrame
 function Base.convert(::Type{DataFrame}, ϕ::Factor)
  df = DataFrames.DataFrame(pattern(ϕ))
  DataFrames.rename!(df, [f => t for (f, t) = zip(names(df), names(ϕ))] )
- df[:potential] = ϕ.potential[:]
+ df[!,:potential] = ϕ.potential[:]
 
  return df
 end

src/gibbs.jl

@@ -329,7 +329,7 @@ function gibbs_sample(bn::BayesNet, nsamples::Integer, burn_in::Integer;
  # for burn_in_initial_sample use get_weighted_dataframe, should be consistent with the varibale consistent_with
  if initial_sample == nothing
  rand_samples = get_weighted_dataframe(bn, 50, consistent_with)
- if reduce(|, isnan.(convert(Array{AbstractFloat}, rand_samples[:p])))
+ if reduce(|, isnan.(convert(Array{AbstractFloat}, rand_samples[!,:p])))
  error("Gibbs Sampler was unable to find an inital sample with non-zero probability, please supply an inital sample")
  end
  burn_in_initial_sample = sample_weighted_dataframe(rand_samples)

src/learning.jl

@@ -3,7 +3,7 @@ function Distributions.fit(::Type{BayesNet}, data::DataFrame, dag::DAG, cpd_type
  length(cpd_types) == nv(dag) || throw(DimensionMismatch("dag and cpd_types must have the same length"))
 
  cpds = Array{CPD}(undef, length(cpd_types))
- tablenames = names(data)
+ tablenames = propertynames(data)
  for (i, target) in enumerate(tablenames)
  C = cpd_types[i]
  parents = tablenames[inneighbors(dag, i)]
@@ -15,7 +15,7 @@ end
 function Distributions.fit(::Type{BayesNet}, data::DataFrame, dag::DAG, ::Type{C}) where {C<:CPD}
 
  cpds = Array{C}(undef, nv(dag))
- tablenames = names(data)
+ tablenames = propertynames(data)
  for (i, target) in enumerate(tablenames)
  parents = tablenames[inneighbors(dag, i)]
  cpds[i] = fit(C, data, target, parents)
@@ -26,7 +26,7 @@ end
 Distributions.fit(::Type{BayesNet{T}}, data::DataFrame, dag::DAG) where {T<:CPD} = fit(BayesNet, data, dag, T)
 
 function _get_dag(data::DataFrame, edges::Tuple{Vararg{Pair{NodeName, NodeName}}})
- varnames = names(data)
+ varnames = propertynames(data)
  dag = DAG(length(varnames))
  for (a,b) in edges
  i = findfirst(isequal(a), varnames)
@@ -109,7 +109,7 @@ As score_component(ScoringFunction, cpd, data), but returns pre-computed values
 if they exist, and populates the cache if they don't
 """
 function _get_parent_indeces(parents::NodeNames, data::DataFrame)
- varnames = names(data)
+ varnames = propertynames(data)
  retval = Array{Int}(undef, length(parents))
  for (i,p) in enumerate(parents)
  retval[i] = something(findfirst(isequal(p), varnames), 0)
@@ -124,7 +124,7 @@ function score_component(
  )
 
  pinds = _get_parent_indeces(parents(cpd), data)
- varnames = names(data)
+ varnames = propertynames(data)
  i = something(findfirst(isequal(name(cpd)), varnames), 0)
 
  if !haskey(cache[i], pinds)

test/test_cpds.jl

@@ -202,6 +202,6 @@ let
  push!(bn, a)
  push!(bn, b)
  push!(bn, c)
- @test mean(rand(bn, 20,:a=>0)[:c]) <= 0.6
- @test mean(rand(bn, 20, :a=>1, :b=>1)[:c]) == 1;
+ @test mean(rand(bn, 20,:a=>0)[!,:c]) <= 0.6
+ @test mean(rand(bn, 20, :a=>1, :b=>1)[!,:c]) == 1;
 end

test/test_factors.jl

@@ -46,7 +46,7 @@ name = :N5
 
 ϕ = Factor(bn, name)
 df = join(DataFrame(ϕ), table(bn, name).potential, on=names(ϕ))
-diff = abs.(df[:p] - df[:potential])
+diff = abs.(df[!,:p] - df[!,:potential])
 
 @test all(diff .< 1E-13)
 end

test/test_gibbs.jl

@@ -1,6 +1,6 @@
 let
 
- # A â~F~R C â~F~P B
+ # A �~F~R C �~F~P B
  bn = BayesNet()
  push!(bn, StaticCPD(:a, Categorical([1.0,0.0])))
  push!(bn, StaticCPD(:b, Categorical([0.0,1.0])))
@@ -10,16 +10,16 @@ let
  variable_order=nothing, time_limit=nothing, 
  error_if_time_out=true, initial_sample=nothing)
  @test size(t5) == (5, 3)
- @test t5[:a] == [1,1,1,1,1]
- @test t5[:b] == [2,2,2,2,2]
- @test t5[:c] == [1,1,1,1,1]
+ @test t5[!,:a] == [1,1,1,1,1]
+ @test t5[!,:b] == [2,2,2,2,2]
+ @test t5[!,:c] == [1,1,1,1,1]
 
  config = GibbsSampler(burn_in=100, thinning=5)
  t5 = rand(bn, config, 5)
  @test size(t5) == (5, 3)
- @test t5[:a] == [1,1,1,1,1]
- @test t5[:b] == [2,2,2,2,2]
- @test t5[:c] == [1,1,1,1,1]
+ @test t5[!,:a] == [1,1,1,1,1]
+ @test t5[!,:b] == [2,2,2,2,2]
+ @test t5[!,:c] == [1,1,1,1,1]
 
  bn2 = BayesNet()
  push!(bn2, StaticCPD(:a, Categorical([0.5,0.5])))
@@ -31,30 +31,30 @@ let
  t6 = gibbs_sample(bn2, 5, 100; thinning=5, consistent_with=Assignment(:c=>1),
  variable_order=nothing, time_limit=nothing,
  error_if_time_out=true, initial_sample=nothing)
- @test t6[:a] == [1,1,1,1,1]
- @test t6[:b] == [1,1,1,1,1]
- @test t6[:c] == [1,1,1,1,1]
+ @test t6[!,:a] == [1,1,1,1,1]
+ @test t6[!,:b] == [1,1,1,1,1]
+ @test t6[!,:c] == [1,1,1,1,1]
 
  config.evidence = Assignment(:c=>1)
  t6 = rand(bn2, config, 5)
- @test t6[:a] == [1,1,1,1,1]
- @test t6[:b] == [1,1,1,1,1]
- @test t6[:c] == [1,1,1,1,1]
+ @test t6[!,:a] == [1,1,1,1,1]
+ @test t6[!,:b] == [1,1,1,1,1]
+ @test t6[!,:c] == [1,1,1,1,1]
 
 
  t7 = gibbs_sample(bn2, 5, 100; thinning=5, consistent_with=Assignment(:c=>2),
  variable_order=nothing, time_limit=nothing,
  error_if_time_out=true, initial_sample=nothing)
- @test t7[:a] == [2,2,2,2,2]
- @test t7[:b] == [1,1,1,1,1]
- @test t7[:c] == [2,2,2,2,2]
+ @test t7[!,:a] == [2,2,2,2,2]
+ @test t7[!,:b] == [1,1,1,1,1]
+ @test t7[!,:c] == [2,2,2,2,2]
 
  rand()
  config.evidence = Assignment(:c=>2)
  t7 = rand(bn2, config, 5)
- @test t7[:a] == [2,2,2,2,2]
- @test t7[:b] == [1,1,1,1,1]
- @test t7[:c] == [2,2,2,2,2]
+ @test t7[!,:a] == [2,2,2,2,2]
+ @test t7[!,:b] == [1,1,1,1,1]
+ @test t7[!,:c] == [2,2,2,2,2]
 
  bn3 = BayesNet()
  push!(bn3, StaticCPD(:a, Normal(2.5, 1.5)))
@@ -107,18 +107,18 @@ let
  variable_order=nothing, time_limit=nothing,
  error_if_time_out=true, initial_sample=nothing)
  @test size(t12) == (5, 3)
- @test t12[:a] == [1,1,1,1,1]
- @test t12[:b] == [2,2,2,2,2]
- @test t12[:c] == [1,1,1,1,1]
+ @test t12[!,:a] == [1,1,1,1,1]
+ @test t12[!,:b] == [2,2,2,2,2]
+ @test t12[!,:c] == [1,1,1,1,1]
 
 
  t13 = gibbs_sample(d_bn, 5, 100; thinning=5, consistent_with=Assignment(:c=>1),
  variable_order=nothing, time_limit=nothing,
  error_if_time_out=true, initial_sample=nothing)
  @test size(t13) == (5, 3)
- @test t13[:a] == [1,1,1,1,1]
- @test t13[:b] == [2,2,2,2,2]
- @test t13[:c] == [1,1,1,1,1]
+ @test t13[!,:a] == [1,1,1,1,1]
+ @test t13[!,:b] == [2,2,2,2,2]
+ @test t13[!,:c] == [1,1,1,1,1]
 
  # test bad parameters given to gibbs_sample
  pass = false

test/test_sampling.jl

@@ -10,8 +10,8 @@ let
 
  t1 = rand(bn, 5)
  @test size(t1) == (5,3)
- @test t1[:a] == [1,1,1,1,1]
- @test t1[:b] == [2,2,2,2,2]
+ @test t1[!,:a] == [1,1,1,1,1]
+ @test t1[!,:b] == [2,2,2,2,2]
 
  t2 = rand(bn, 5, Assignment(:c=>1))
  @test size(t1) == (5,3)
@@ -23,6 +23,6 @@ let
  # is there a test here?
 
  t5 = rand(bn, GibbsSampler(Assignment(:c=>1), burn_in=5), 5)
- @test t5[:a] == [1,1,1,1,1]
- @test t5[:b] == [2,2,2,2,2]
+ @test t5[!,:a] == [1,1,1,1,1]
+ @test t5[!,:b] == [2,2,2,2,2]
 end