Fix #41, add tests around evaluate/tuning in MLJ

Project.toml

@@ -4,13 +4,15 @@ authors = ["Beacon Biosignals, Inc."]
 version = "0.3.6"
 
 [deps]
+CategoricalArrays = "324d7699-5711-5eae-9e2f-1d82baa6b597"
 MLJModelInterface = "e80e1ace-859a-464e-9ed9-23947d8ae3ea"
 OrderedCollections = "bac558e1-5e72-5ebc-8fee-abe8a469f55d"
 PythonCall = "6099a3de-0909-46bc-b1f4-468b9a2dfc0d"
 Tables = "bd369af6-aec1-5ad0-b16a-f7cc5008161c"
 
 [compat]
 Aqua = "0.8.4"
+CategoricalArrays = "0.10.8"
 DataFrames = "1.6"
 MLJBase = "1"
 MLJModelInterface = "1.7"

src/MLJCatBoostInterface.jl

@@ -9,6 +9,7 @@ using Tables
 using MLJModelInterface: MLJModelInterface
 const MMI = MLJModelInterface
 using MLJModelInterface: Table, Continuous, Count, Finite, OrderedFactor, Multiclass
+using CategoricalArrays: CategoricalArray, CategoricalValue
 const PKG = "CatBoost"
 
 """

src/mlj_catboostclassifier.jl

@@ -72,8 +72,12 @@ function model_init(mlj_model::CatBoostClassifier; kw...)
     return catboost.CatBoostClassifier(; mlj_to_kwargs(mlj_model)..., kw...)
 end
 
+prepare_single_class(y_first) = CategoricalArray([y_first])[1]
+prepare_single_class(y_first::CategoricalValue) = y_first
+
 function MMI.fit(mlj_model::CatBoostClassifier, verbosity::Int, data_pool, y_first)
-    # Check if y_first has only one unique value
+    # Check if the data pool has only one unique value
+    y_first = prepare_single_class(y_first)
     unique_classes = pyconvert(Vector, numpy.unique(data_pool.get_label()))
     if length(unique_classes) == 1
         # Skip training and store the single class
@@ -103,8 +107,7 @@ function MMI.predict(mlj_model::CatBoostClassifier, fitresult, X_pool)
         n = pyconvert(Int, X_pool.shape[0])
         classes = [fitresult.single_class]
         probs = ones(n, 1)
-        pool = MMI.categorical([fitresult.y_first])
-        return MMI.UnivariateFinite(classes, probs; pool=pool)
+        return MMI.UnivariateFinite(classes, probs; pool=fitresult.y_first)
     end
 
     model, y_first = fitresult