Fix multi-target MLJFlux example

.gitignore

@@ -1,9 +1,11 @@
 /Manifest.toml
 /docs/Manifest.toml
+/test/Manifest.toml
 .ipynb_checkpoints
 *~
 #*
 *.bu
 .DS_Store
 sandbox/
-docs/build
+docs/build
+.vscode/

src/utilities.jl

@@ -6,7 +6,7 @@ function plotting_report(fields, scales, history)
     n_parameters = length(fields)
 
     A = Array{Any}(undef, (n_models, n_parameters))
-    measurements = Vector{Float64}(undef, n_models)
+    measurements = Vector{eltype(first(history).measurement)}(undef, n_models)
 
     for j in eachindex(history)
         entry = history[j]

test/Project.toml

@@ -5,9 +5,11 @@ DecisionTree = "7806a523-6efd-50cb-b5f6-3fa6f1930dbb"
 Distances = "b4f34e82-e78d-54a5-968a-f98e89d6e8f7"
 Distributed = "8ba89e20-285c-5b6f-9357-94700520ee1b"
 Distributions = "31c24e10-a181-5473-b8eb-7969acd0382f"
+Flux = "587475ba-b771-5e3f-ad9e-33799f191a9c"
 LatinHypercubeSampling = "a5e1c1ea-c99a-51d3-a14d-a9a37257b02d"
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
 MLJBase = "a7f614a8-145f-11e9-1d2a-a57a1082229d"
+MLJFlux = "094fc8d1-fd35-5302-93ea-dabda2abf845"
 MLJModelInterface = "e80e1ace-859a-464e-9ed9-23947d8ae3ea"
 MultivariateStats = "6f286f6a-111f-5878-ab1e-185364afe411"
 NearestNeighbors = "b8a86587-4115-5ab1-83bc-aa920d37bbce"
@@ -26,7 +28,9 @@ ComputationalResources = "0.3"
 DecisionTree = "0.10"
 Distances = "0.10"
 Distributions = "0.25"
+Flux = "0.13"
 MLJBase = "0.20"
+MLJFlux = "0.2"
 MLJModelInterface = "1.3"
 MultivariateStats = "0.9"
 NearestNeighbors = "0.4"

test/models.jl

@@ -13,6 +13,7 @@ import MLJModelInterface
 include("models/Constant.jl")
 include("models/DecisionTree.jl")
 include("models/NearestNeighbors.jl")
+include("models/Flux.jl")
 include("models/MultivariateStats.jl")
 include("models/Transformers.jl")
 include("models/foobarmodel.jl")

test/models/Flux.jl

@@ -0,0 +1,35 @@
+using MLJFlux, MLJ, Flux
+
+X = randn(100, 2)
+Y = X * rand(2, 2) .+ 0.1 * randn.()
+XT = MLJ.table(X, names = [:x1, :x2])
+YT = MLJ.table(Y, names = [:y1, :y2])
+act = tanh
+nn = Chain(
+  Dense(2, 5, act),
+  Dense(5, 5, act),
+  Dense(5, 5, act),
+  Dense(5, 2, identity),
+)
+builder = MLJFlux.@builder nn
+
+function multi_target(loss)
+  (x1, x2) -> sum(map(x1, x2) do _x1, _x2
+    loss(_x1, _x2)
+  end)
+end
+loss = multi_target(l2)
+model = MLJFlux.MultitargetNeuralNetworkRegressor(builder = builder; epochs = 10, loss)
+r = (MLJ.range(model, :lambda, lower=1e-6, upper=1.0, scale=:linear), 10)
+tuning = MLJ.Grid(shuffle = true)
+tuned_model = MLJ.TunedModel(
+  model;
+  tuning,
+  resampling = MLJ.CV(nfolds = 5),
+  range = [r],
+  measure = loss,
+  n = 10,
+  check_measure = false,
+)
+mach = MLJ.machine(tuned_model, XT, YT)
+MLJ.fit!(mach, verbosity=1)