Fix and add rawmodel

src/interfaces/MLJ.jl

@@ -100,8 +100,11 @@ function MMI.fit(m::SymbolicModel, verbosity::Integer, X, y, var_grouping, class
         # syntaxstring_kwargs = (; hidemodality = (length(var_grouping) == 1), variable_names_map = var_grouping)
     ))
 
+    rawmodel_full = model
+    rawmodel = MDT.prune(model; simplify = true)
+
     solemodel_full = translate_function(model)
-    solemodel = translate_function(MDT.prune(model; simplify = true))
+    solemodel = translate_function(rawmodel)
 
     fitresult = (
         model         = model,
@@ -115,9 +118,11 @@ function MMI.fit(m::SymbolicModel, verbosity::Integer, X, y, var_grouping, class
             (Xnew, ynew, var_grouping, classes_seen, w) = MMI.reformat(m, Xnew, ynew; passive_mode = true)
             translate_function(ModalDecisionTrees.sprinkle(model, Xnew, ynew))
         end,
-        model                       = model,
-        model_full                  = model_full,
         # TODO remove redundancy?
+        model                       = solemodel,
+        model_full                  = solemodel_full,
+        rawmodel                    = rawmodel,
+        rawmodel_full               = rawmodel_full,
         solemodel                   = solemodel,
         solemodel_full              = solemodel_full,
         var_grouping                = var_grouping,
@@ -135,8 +140,8 @@ function MMI.fit(m::SymbolicModel, verbosity::Integer, X, y, var_grouping, class
     return fitresult, cache, report
 end
 
-MMI.fitted_params(::TreeModel, fitresult) = merge(fitresult, (; tree = fitresult.model))
-MMI.fitted_params(::ForestModel, fitresult) = merge(fitresult, (; forest = fitresult.model))
+MMI.fitted_params(::TreeModel, fitresult) = merge(fitresult, (; tree = fitresult.rawmodel))
+MMI.fitted_params(::ForestModel, fitresult) = merge(fitresult, (; forest = fitresult.rawmodel))
 
 ############################################################################################
 ############################################################################################
@@ -149,7 +154,7 @@ function MMI.predict(m::SymbolicModel, fitresult, Xnew, var_grouping = nothing)
             "var_grouping = $(var_grouping)" *
             "\n"
     end
-    MDT.apply_proba(fitresult.model, Xnew, get(fitresult, :classes_seen, nothing); suppress_parity_warning = true)
+    MDT.apply_proba(fitresult.rawmodel, Xnew, get(fitresult, :classes_seen, nothing); suppress_parity_warning = true)
 end
 
 ############################################################################################
@@ -187,7 +192,7 @@ function MMI.feature_importances(m::SymbolicModel, fitresult, report)
         error("Unexpected feature_importance encountered: $(m.feature_importance).")
     end
 
-    featimportance_dict = compute_featureimportance(fitresult.model, fitresult.var_grouping; normalize=true)
+    featimportance_dict = compute_featureimportance(fitresult.rawmodel, fitresult.var_grouping; normalize=true)
     featimportance_vec = collect(featimportance_dict)
     sort!(featimportance_vec, rev=true, by=x->last(x))
 

src/interfaces/MLJ/printer.jl

@@ -29,7 +29,7 @@ function (c::ModelPrinter)(
     max_depth::Union{Nothing,Integer} = c.m.display_depth;
     kwargs...
 )
-    c(io, (print_solemodel ? c.solemodel : c.model); max_depth = max_depth, kwargs...)
+    c(io, (print_solemodel ? c.solemodel : c.rawmodel); max_depth = max_depth, kwargs...)
 end
 
 function (c::ModelPrinter)(

test/classification/demo-juliacon2022.jl

@@ -62,6 +62,6 @@ cm = ConfusionMatrix(y_test, y_test_preds; force_class_order=["I have command",
 @test overall_accuracy(cm) > 0.6
 
 # Render model in LaTeX
-# show_latex(mach.fitresult.model; variable_names = [variable_names_latex], silent = true);
+# show_latex(mach.fitresult.rawmodel; variable_names = [variable_names_latex], silent = true);
 
 end

test/classification/digits.jl

@@ -18,13 +18,13 @@ train_idxs, test_idxs = p[1:round(Int, N*.4)], p[round(Int, N*.4)+1:end]
 
 # Full training
 mach = machine(ModalDecisionTree(;), X, y) |> fit!
-@test nnodes(fitted_params(mach).model) == 191
+@test nnodes(fitted_params(mach).rawmodel) == 191
 @test sum(predict_mode(mach, X) .== y) / length(y) > 0.92
 
 ############################################################################################
 
 mach = machine(ModalDecisionTree(;), X, y) |> m->fit!(m, rows = train_idxs)
-@test nnodes(fitted_params(mach).model) == 115
+@test nnodes(fitted_params(mach).rawmodel) == 115
 @test sum(predict_mode(mach, rows = test_idxs) .== y[test_idxs]) / length(y[test_idxs]) > 0.78
 
 
@@ -33,7 +33,7 @@ mach = machine(ModalDecisionTree(;
     max_depth           = 6,
     min_samples_leaf    = 5,
 ), X, y) |> m->fit!(m, rows = train_idxs)
-@test nnodes(fitted_params(mach).model) == 77
+@test nnodes(fitted_params(mach).rawmodel) == 77
 @test sum(predict_mode(mach, rows = test_idxs) .== y[test_idxs]) / length(y[test_idxs]) > 0.75
 
 
@@ -47,7 +47,7 @@ mach = machine(ModalRandomForest(;
     min_purity_increase = 0.0,
     rng = Random.MersenneTwister(1)
 ), X, y) |> m->fit!(m, rows = train_idxs)
-@test nnodes(fitted_params(mach).model) == 1242
+@test nnodes(fitted_params(mach).rawmodel) == 1242
 @test predict_mode(mach, rows = test_idxs)
 @test predict(mach, rows = test_idxs)
 @test sum(predict_mode(mach, rows = test_idxs) .== y[test_idxs]) / length(y[test_idxs]) > 0.85
@@ -57,15 +57,15 @@ mach = machine(ModalRandomForest(;
 
 # NamedTuple dataset
 mach = mach = machine(ModalDecisionTree(;), Xnt, y) |> m->fit!(m, rows = train_idxs)
-@test nnodes(fitted_params(mach).model) == 131
+@test nnodes(fitted_params(mach).rawmodel) == 131
 @test sum(predict_mode(mach, rows = test_idxs) .== y[test_idxs]) / length(y[test_idxs]) > 0.68
 
 mach = machine(ModalDecisionTree(;
     relations = :IA7,
     features = [minimum],
     initconditions = :start_at_center,
 ), Xnt, y) |> m->fit!(m, rows = train_idxs)
-@test nnodes(fitted_params(mach).model) == 147
+@test nnodes(fitted_params(mach).rawmodel) == 147
 @test sum(predict_mode(mach, rows = test_idxs) .== y[test_idxs]) / length(y[test_idxs]) > 0.67
 
 mach = machine(ModalDecisionTree(;
@@ -74,7 +74,7 @@ mach = machine(ModalDecisionTree(;
     # initconditions = :start_at_center,
     featvaltype = Float32,
 ), selectrows(Xnt, train_idxs), selectrows(y, train_idxs)) |> m->fit!(m)
-@test nnodes(fitted_params(mach).model) == 131
+@test nnodes(fitted_params(mach).rawmodel) == 131
 @test sum(predict_mode(mach, selectrows(Xnt, test_idxs)) .== y[test_idxs]) / length(y[test_idxs]) > 0.71
 
 
@@ -111,7 +111,7 @@ mach = machine(ModalDecisionTree(;
     # initconditions = :start_at_center,
     featvaltype = Float32,
 ), selectrows(Xnt, train_idxs), selectrows(y, train_idxs)) |> m->fit!(m)
-@test nnodes(fitted_params(mach).model) == 71
+@test nnodes(fitted_params(mach).rawmodel) == 71
 @test sum(predict_mode(mach, selectrows(Xnt, test_idxs)) .== y[test_idxs]) / length(y[test_idxs]) > 0.73
 
 preds, tree2 = report(mach).sprinkle(selectrows(Xnt, test_idxs), selectrows(y, test_idxs));
@@ -129,7 +129,7 @@ mach = machine(ModalDecisionTree(;
     # initconditions = :start_at_center,
     featvaltype = Float32,
 ), selectrows(Xnt, train_idxs), selectrows(y, train_idxs)) |> m->fit!(m)
-@test nnodes(fitted_params(mach).model) == 79
+@test nnodes(fitted_params(mach).rawmodel) == 79
 @test sum(predict_mode(mach, selectrows(Xnt, test_idxs)) .== y[test_idxs]) / length(y[test_idxs]) > 0.73
 
 preds, tree2 = report(mach).sprinkle(selectrows(Xnt, test_idxs), selectrows(y, test_idxs));
@@ -141,15 +141,15 @@ printmodel.(joinrules(listrules(ModalDecisionTrees.translate(tree2))); show_metr
 readmetrics.(joinrules(listrules(ModalDecisionTrees.translate(tree2))))
 
 mach = machine(ModalDecisionTree(;), Xnt, y) |> m->fit!(m, rows = train_idxs)
-@test nnodes(fitted_params(mach).model) == 137
+@test nnodes(fitted_params(mach).rawmodel) == 137
 @test sum(predict_mode(mach, rows = test_idxs) .== y[test_idxs]) / length(y[test_idxs]) > 0.70
 
 mach = machine(ModalDecisionTree(;
     n_subfeatures       = 0,
     max_depth           = 6,
     min_samples_leaf    = 5,
 ), Xnt, y) |> m->fit!(m, rows = train_idxs)
-@test nnodes(fitted_params(mach).model) == 77
+@test nnodes(fitted_params(mach).rawmodel) == 77
 @test sum(predict_mode(mach, rows = test_idxs) .== y[test_idxs]) / length(y[test_idxs]) > 0.75
 
 
@@ -162,5 +162,5 @@ mach = machine(ModalRandomForest(;
     min_samples_split   = 2,
     min_purity_increase = 0.0,
 ), Xnt, y) |> m->fit!(m, rows = train_idxs)
-@test nnodes(fitted_params(mach).model) == 77
+@test nnodes(fitted_params(mach).rawmodel) == 77
 @test sum(predict_mode(mach, rows = test_idxs) .== y[test_idxs]) / length(y[test_idxs]) > 0.75

test/classification/iris-params.jl

@@ -7,12 +7,12 @@ X, y = @load_iris
 
 model = ModalDecisionTree(; max_depth = 0)
 mach = machine(model, X, y) |> fit!
-@test height(fitted_params(mach).model) == 0
-@test depth(fitted_params(mach).model) == 0
+@test height(fitted_params(mach).rawmodel) == 0
+@test depth(fitted_params(mach).rawmodel) == 0
 
 model = ModalDecisionTree(; max_depth = 2, )
 mach = machine(model, X, y) |> fit!
-@test depth(fitted_params(mach).model) == 2
+@test depth(fitted_params(mach).rawmodel) == 2
 
 model = ModalDecisionTree(;
 	min_samples_leaf = 2,

test/classification/iris.jl

@@ -24,11 +24,11 @@ yhat = MLJ.predict_mode(mach, X)
 
 @test MLJBase.accuracy(y, yhat) > 0.8
 
-@test_nowarn fitted_params(mach).model
+@test_nowarn fitted_params(mach).rawmodel
 @test_nowarn report(mach).solemodel
 
-@test_nowarn printmodel(prune(fitted_params(mach).model, simplify=true, min_samples_leaf = 20), max_depth = 3)
-@test_nowarn printmodel(prune(fitted_params(mach).model, simplify=true, min_samples_leaf = 20))
+@test_nowarn printmodel(prune(fitted_params(mach).rawmodel, simplify=true, min_samples_leaf = 20), max_depth = 3)
+@test_nowarn printmodel(prune(fitted_params(mach).rawmodel, simplify=true, min_samples_leaf = 20))
 
 @test_nowarn printmodel(report(mach).solemodel, header = false)
 @test_nowarn printmodel(report(mach).solemodel, header = :brief)

test/classification/japanesevowels.jl

@@ -54,7 +54,7 @@ acc = sum(yhat .== y[test_idxs])/length(yhat)
 @test_throws ErrorException listrules(report(mach).solemodel; use_shortforms=false, use_leftmostlinearform = true, force_syntaxtree = true)
 
 # Access raw model
-fitted_params(mach).model;
+fitted_params(mach).rawmodel;
 report(mach).printmodel(3);
 
 MLJ.fit!(mach)
@@ -113,5 +113,5 @@ yhat = MLJ.predict_mode(mach, rows=test_idxs)
 acc = sum(yhat .== y[test_idxs])/length(yhat)
 MLJ.kappa(yhat, y[test_idxs])
 
-@test_nowarn prune(fitted_params(mach).model, simplify=true)
-@test_nowarn prune(fitted_params(mach).model, simplify=true, min_samples_leaf = 20)
+@test_nowarn prune(fitted_params(mach).rawmodel, simplify=true)
+@test_nowarn prune(fitted_params(mach).rawmodel, simplify=true, min_samples_leaf = 20)

test/other-test-stuff.jl

@@ -44,7 +44,7 @@ fitresult = MMI.fit(model, 0, X_train, Y_train);
 
 Y_test_preds, test_tree = MMI.predict(model, fitresult[1], X_test, Y_test);
 
-tree = fitresult[1].model
+tree = fitresult[1].rawmodel
 
 fitresult[3].print_tree()
 
@@ -57,7 +57,7 @@ println(test_tree)
 # SoleModels.ConfusionMatrix(Y_test_preds, Y_test)
 
 
-# tree = fitresult.model
+# tree = fitresult.rawmodel
 # println(tree)
 # println(test_tree)
 # fitreport.print_tree()
@@ -99,7 +99,7 @@ show_latex(test_tree, "test", [variable_names_latex])
 
 #     Y_test_preds, test_tree = MMI.predict(model, fitresult[1], X_static_test, Y_test);
 
-#     tree = fitresult[1].model
+#     tree = fitresult[1].rawmodel
 
 #     fitresult[3].print_tree()