add feature importances support for tuned models

src/tuned_models.jl

@@ -856,6 +856,22 @@ function MLJBase.training_losses(tuned_model::EitherTunedModel, _report)
     return ret
 end
 
+## Support for Feature Importances
+function MLJBase.reports_feature_importances(::Type{<:EitherTunedModel{<:Any,M}}) where {M}
+    return MLJBase.reports_feature_importances(M)
+end
+
+function MLJBase.reports_feature_importances(model::EitherTunedModel)
+    return MLJBase.reports_feature_importances(model.model)
+end # This is needed in some cases (e.g tuning a `Pipeline`)
+
+function MLJBase.feature_importances(::EitherTunedModel, fitresult, report)
+    # fitresult here is a machine created using the best_model obtained 
+    # from the tuning process.
+    # The line below will return `nothing` when the model being tuned doesn't 
+    # support feature_importances.
+    return MLJBase.feature_importances(fitresult)
+end
 
 ## METADATA
 

test/models/DecisionTree.jl

@@ -95,8 +95,11 @@ function MLJBase.fit(model::DecisionTreeClassifier, verbosity::Int, X, y)
     #> empty values):
 
     cache = nothing
-    report = (classes_seen=classes_seen,
-              print_tree=TreePrinter(tree))
+    report = (
+        classes_seen=classes_seen,
+        print_tree=TreePrinter(tree),
+        features=collect(Tables.columnnames(Tables.columns(X)))
+    )
 
     return fitresult, cache, report
 end
@@ -134,6 +137,17 @@ function MLJBase.predict(model::DecisionTreeClassifier
             for i in 1:size(y_probabilities, 1)]
 end
 
+MLJBase.reports_feature_importances(::Type{<:DecisionTreeClassifier}) = true
+
+function MMI.feature_importances(m::DecisionTreeClassifier, fitresult, report)
+    features = report.features
+    fi = DecisionTree.impurity_importance(first(fitresult), normalize=true)
+    fi_pairs = Pair.(features, fi)
+    # sort descending
+    sort!(fi_pairs, by= x->-x[2])
+
+    return fi_pairs
+end
 
 ## REGRESSOR
 

test/tuned_models.jl

@@ -13,19 +13,24 @@ using Random
 Random.seed!(1234*myid())
 using .TestUtilities
 
-N = 30
-x1 = rand(N);
-x2 = rand(N);
-x3 = rand(N);
-X = (; x1, x2, x3);
-y = 2*x1 .+ 5*x2 .- 3*x3 .+ 0.4*rand(N);
-
-m(K) = KNNRegressor(; K)
-r = [m(K) for K in 13:-1:2]
-
-# TODO: replace the above with the line below and post an issue on
-# the failure (a bug in Distributed, I reckon):
-# r = (m(K) for K in 13:-1:2)
+begin 
+    N = 30
+    x1 = rand(N);
+    x2 = rand(N);
+    x3 = rand(N);
+    X = (; x1, x2, x3);
+    y = 2*x1 .+ 5*x2 .- 3*x3 .+ 0.4*rand(N);
+
+    m(K) = KNNRegressor(; K)
+    r = [m(K) for K in 13:-1:2]
+
+    Xtree, yhat = @load_iris
+    trees = [DecisionTreeClassifier(pruning_purity = rand()) for _ in 13:-1:2]
+
+    # TODO: replace the above with the line below and post an issue on
+    # the failure (a bug in Distributed, I reckon):
+    # r = (m(K) for K in 13:-1:2)
+end
 
 @testset "constructor" begin
     @test_throws(MLJTuning.ERR_SPECIFY_RANGE,
@@ -105,6 +110,10 @@ end
     @test _report.best_model == collect(r)[best_index]
     @test _report.history[5] == MLJTuning.delete(history[5], :metadata)
 
+    # feature_importances:
+    # This should return nothing as `KNNRegressor` doesn't support feature_importances
+    @test feature_importances(tm, fitresult, _report) === nothing
+
     # training_losses:
     losses = training_losses(tm, _report)
     @test all(eachindex(losses)) do i
@@ -146,6 +155,22 @@ end
     @test results4 ≈ results
 end
 
+@testset_accelerated "Feature Importances" accel begin
+    # the DecisionTreeClassifier in /test/_models/ supports feature importances.
+    tm = TunedModel(
+        models = trees,
+        resampling = CV(nfolds=2),
+        measures = cross_entropy,
+        acceleration = CPU1(),
+        acceleration_resampling = accel
+    )
+    @test reports_feature_importances(tm)
+    fitresult, _, report = MLJBase.fit(tm, 0, Xtree, yhat)
+    features = first.(feature_importances(tm, fitresult, report))
+    @test Set(features) == Set(keys(Xtree))
+
+end
+
 @testset_accelerated(
     "under/over supply of models",
     accel,