do not require DataFrames.jl when output doesn't contain DataFrames

src/datasets/misc/boston_housing.jl

@@ -77,21 +77,24 @@ end
 function BostonHousing(; as_df = true, dir = nothing)
     @assert dir === nothing "custom `dir` is not supported at the moment."
     path = joinpath(@__DIR__, "..", "..", "..", "data", "boston_housing.csv")
-    df = read_csv(path)
-    features = df[!, DataFrames.Not(:MEDV)]
-    targets = df[!, [:MEDV]]
+    t = read_csv(path, CSV.File)
+    colnames = Tables.columnnames(t)
+    features = table_to_matrix(t, select = colnames[1:end-1])
+    targets = table_to_matrix(t, select = colnames[end:end])
+
 
     metadata = Dict{String, Any}()
     metadata["path"] = path
-    metadata["feature_names"] = names(features)
-    metadata["target_names"] = names(targets)
-    metadata["n_observations"] = size(targets, 1)
+    metadata["feature_names"] = colnames[1:end-1]
+    metadata["target_names"] = colnames[end:end]
+    metadata["n_observations"] = size(features, 1)
     metadata["description"] = BOSTONHOUSING_DESCR
 
-    if !as_df 
-        features = df_to_matrix(features) 
-        targets = df_to_matrix(targets)
-        df = nothing
+    df = nothing
+    if as_df
+        df = table_to_df(t, names = colnames)
+        features = matrix_to_df(features, names = colnames[1:end-1])
+        targets = matrix_to_df(targets, names = colnames[end:end])
     end
 
     return BostonHousing(metadata, features, targets, df)

src/datasets/misc/iris.jl

@@ -79,22 +79,23 @@ end
 
 function Iris(; dir = nothing, as_df = true)
     path = datafile("Iris", "iris.data", dir)
-    df = read_csv(path, header=0)
-    DataFrames.rename!(df, ["sepallength", "sepalwidth", "petallength", "petalwidth", "class"])
-
-    features = df[!, DataFrames.Not(:class)]
-    targets = df[!, [:class]]
-
+    t = read_csv(path, CSV.File, header=0)
+    colnames = Tables.columnnames(t)
+    truecolnames = ["sepallength", "sepalwidth", "petallength", "petalwidth", "class"]
+    features = table_to_matrix(t, select = colnames[1:end-1])
+    targets = table_to_matrix(t, select = colnames[end:end])
+    
     metadata = Dict{String, Any}()
     metadata["path"] = path
-    metadata["n_observations"] = size(df, 1)
-    metadata["feature_names"] = names(features)
-    metadata["target_names"] = names(targets)
-
-    if !as_df
-        features = df_to_matrix(features)
-        targets = df_to_matrix(targets)
-        df = nothing
+    metadata["n_observations"] = size(features, 1)
+    metadata["feature_names"] = truecolnames[1:end-1]
+    metadata["target_names"] = truecolnames[end:end]
+
+    df = nothing
+    if as_df
+        df = table_to_df(t, names = truecolnames)
+        features = matrix_to_df(features, names = truecolnames[1:end-1])
+        targets = matrix_to_df(targets, names = truecolnames[end:end])
     end
 
     return Iris(metadata, features, targets, df)

src/datasets/misc/titanic.jl

@@ -61,22 +61,25 @@ end
 function Titanic(; as_df = true, dir = nothing)
     @assert dir === nothing "custom `dir` is not supported at the moment."
     path = joinpath(@__DIR__, "..", "..", "..", "data", "titanic.csv")
-    df = read_csv(path)
-
-    features = df[!, DataFrames.Not(:Survived)]
-    targets = df[!, [:Survived]]
+    t = read_csv(path, CSV.File)
+    colnames = Tables.columnnames(t)
+    ncols = length(colnames)
+    # :Sruvived is the second column
+    features = table_to_matrix(t, select = colnames[[1; 3:ncols]])
+    targets = table_to_matrix(t, select = colnames[2:2])
 
     metadata = Dict{String, Any}()
     metadata["path"] = path
-    metadata["feature_names"] = names(features)
-    metadata["target_names"] = names(targets)
-    metadata["n_observations"] = size(df, 1)
+    metadata["feature_names"] = colnames[[1; 3:ncols]]
+    metadata["target_names"] = colnames[2:2]
+    metadata["n_observations"] = size(features, 1)
     metadata["description"] = TITANIC_DESCR
 
-    if !as_df
-        features = df_to_matrix(features)
-        targets = df_to_matrix(targets)
-        df = nothing
+    df = nothing
+    if as_df
+        df = table_to_df(t, names = colnames)
+        features = matrix_to_df(features, names = colnames[[1; 3:ncols]])
+        targets = matrix_to_df(targets, names = colnames[2:2])
     end
 
     return Titanic(metadata, features, targets, df)

src/io.jl

@@ -3,13 +3,12 @@ function read_csv(path; kws...)
     return read_csv_asdf(path; kws...)
 end
 
-# function read_csv(path, sink::Type{<:AbstractMatrix{T}}; delim=nothing, kws...) where T
-#     x = delim === nothing ? readdlm(path, T; kws...) : readdlm(path, delim, T; kws...)
-#     return x
-# end
-
 function read_csv(path, sink::Type{A}; kws...) where A <: AbstractMatrix
-    return A(read_csv(path; kws...))
+    return table_to_matrix(read_csv(path, CSV.File; kws...))
+end
+
+function read_csv(path, sink::Type{CSV.File}; kws...)
+    return CSV.File(path; kws...)
 end
 
 function read_csv_asdf(path; kws...)

src/utils.jl

@@ -21,6 +21,31 @@ function restrict_array_type(res::AbstractArray)
     end
 end
 
+function table_to_matrix(t; select = nothing)
+    if select === nothing
+        cnames = Tables.columnnames(t)
+    else
+        cnames = select
+    end
+    return hcat((Tables.getcolumn(t, n) for n in cnames)...)
+end
+
+function table_to_df(t; names = nothing)
+    df = DataFrames.DataFrame(t)
+    if names !== nothing
+        DataFrames.rename!(df, names)
+    end
+    return df
+end
+
+function matrix_to_df(a::AbstractMatrix; names = nothing)
+    df = DataFrames.DataFrame(a, :auto)
+    if names !== nothing
+        DataFrames.rename!(df, names)
+    end
+    return df
+end
+
 function df_to_matrix(df)
     x = Matrix(df)
     if size(x, 2) == 1

test/test_utils.jl

@@ -29,12 +29,11 @@ function test_inmemory_supervised_table_dataset(d::D;
         @test size(d.features) == (n_obs, n_features)
         @test size(d.targets) == (n_obs, n_targets)
 
-        # check that dataframe shares the same storage of features and targets
         for c in names(d.dataframe)
             if c in names(d.targets)
-                @test d.dataframe[!, c] === d.targets[!,c] 
+                @test d.dataframe[!, c] == d.targets[!,c] 
             else
-                @test d.dataframe[!, c] === d.features[!,c]
+                @test d.dataframe[!, c] == d.features[!,c]
             end
         end