adapt to changes in levels behaviour in CategoricalDistributions

Author: ablaom

src/models/discriminant_analysis.jl

@@ -20,7 +20,8 @@ const ERR_LONE_TARGET_CLASS = ArgumentError(
 )
 
 function _check_lda_data(model, X, y)
-    pool = MMI.classes(y[1]) # Class list containing entries in pool of `y`.
+    pool = CategoricalDistributions.levels(y[1]) # Class list containing entries in pool
+                                                  # of `y`.
     classes_seen = unique(y) # Class list of actual entries in seen in `y`.
     nc = length(classes_seen) # Number of actual classes seen in `y`.
 
@@ -109,7 +110,7 @@ function MMI.predict(m::LDA, (core_res, classes_seen, pool), Xnew)
     Pr .*= -1
     # apply a softmax transformation
     softmax!(Pr)
-    return MMI.UnivariateFinite(classes_seen, Pr, pool=pool)
+    return MMI.UnivariateFinite(classes_seen, Pr)
 end
 
 metadata_model(
@@ -160,7 +161,7 @@ function _check_prob01(priors)
 end
 
 @inline function _check_lda_priors(priors::UnivariateFinite, classes_seen, pool)
-    if MMI.classes(priors) != pool
+    if CategoricalDistributions.levels(priors) != pool
         throw(
             ArgumentError(
                 "UnivariateFinite `priors` must have common pool with training target."
@@ -236,7 +237,7 @@ function MMI.fitted_params(::BayesianLDA, (core_res, classes_seen, pool,  priors
     return (
         classes = classes_seen,
         projection_matrix=MS.projection(core_res),
-        priors=MMI.UnivariateFinite(classes_seen, priors, pool=pool)
+        priors=MMI.UnivariateFinite(classes_seen, priors)
     )
 end
 
@@ -261,7 +262,7 @@ function MMI.predict(m::BayesianLDA, (core_res, classes_seen, pool, priors, n),
 
     # apply a softmax transformation to convert Pr to a probability matrix
     softmax!(Pr)
-    return MMI.UnivariateFinite(classes_seen, Pr, pool=pool)
+    return MMI.UnivariateFinite(classes_seen, Pr)
 end
 
 function MMI.transform(m::T, (core_res, ), X) where T<:Union{LDA, BayesianLDA}
@@ -353,7 +354,7 @@ function MMI.predict(m::SubspaceLDA, (core_res, outdim, classes_seen, pool), Xne
     Pr .*= -1
     # apply a softmax transformation
     softmax!(Pr)
-    return MMI.UnivariateFinite(classes_seen, Pr, pool=pool)
+    return MMI.UnivariateFinite(classes_seen, Pr)
 end
 
 metadata_model(
@@ -430,7 +431,7 @@ function MMI.fitted_params(
   return (
       classes = classes_seen,
       projection_matrix=core_res.projw * view(core_res.projLDA, :, 1:outdim),
-      priors=MMI.UnivariateFinite(classes_seen, priors, pool=pool)
+      priors=MMI.UnivariateFinite(classes_seen, priors)
   )
 end
 
@@ -470,7 +471,7 @@ function MMI.predict(
 
     # apply a softmax transformation to convert Pr to a probability matrix
     softmax!(Pr)
-    return MMI.UnivariateFinite(classes_seen, Pr, pool=pool)
+    return MMI.UnivariateFinite(classes_seen, Pr)
 end
 
 function MMI.transform(
@@ -724,7 +725,8 @@ The fields of `fitted_params(mach)` are:
   section below).
 
 - `priors`: The class priors for classification. As inferred from training target `y`, if
-  not user-specified. A `UnivariateFinite` object with levels consistent with `levels(y)`.
+  not user-specified. A `UnivariateFinite` object with levels (classes) consistent with
+  `levels(y)`.
 
 # Report
 
@@ -954,7 +956,8 @@ The fields of `fitted_params(mach)` are:
   section below).
 
 - `priors`: The class priors for classification. As inferred from training target `y`, if
-  not user-specified. A `UnivariateFinite` object with levels consistent with `levels(y)`.
+  not user-specified. A `UnivariateFinite` object with levels (classes) consistent with
+  `levels(y)`.
 
 # Report
 

test/models/discriminant_analysis.jl

@@ -73,15 +73,15 @@ end
     ytest = selectrows(y, test)
 
     BLDA_model = BayesianLDA(regcoef=0)
-    
+
     ## Check model `fit`
     fitresult, cache, report = fit(BLDA_model, 1, Xtrain, ytrain)
     classes_seen, projection_matrix, priors = fitted_params(BLDA_model, fitresult)
-    @test classes(priors) == classes(y)
+    @test levels(priors) == levels(y)
     @test pdf.(priors, support(priors)) == [491/998, 507/998]
     @test classes_seen == ["Up", "Down"]
     @test round.((report.class_means)', sigdigits = 3) == [-0.0395 -0.0313; 0.0428 0.0339] #[0.0428 0.0339; -0.0395 -0.0313]
-    
+
     ## Check model `predict`
     preds = predict(BLDA_model, fitresult, Xtest)
     mce = cross_entropy(preds, ytest) |> mean
@@ -94,7 +94,7 @@ end
     fitresult1, cache1, report1 = fit(BLDA_model1, 1, Xtrain, ytrain)
     classes_seen1, projection_matrix1, priors1 = fitted_params(BLDA_model1, fitresult1)
     BLDA_model2 = BayesianLDA(
-        regcoef=0, priors=UnivariateFinite(classes(ytrain), [491/998, 507/998])
+        regcoef=0, priors=UnivariateFinite(levels(ytrain), [491/998, 507/998])
     )
     fitresult2, cache2, report2 = fit(BLDA_model2, 1, Xtrain, ytrain)
     classes_seen2, projection_matrix2, priors2 = fitted_params(BLDA_model2, fitresult2)
@@ -156,7 +156,7 @@ end
         LDA_model1, fitresult1
     )
     LDA_model2 = BayesianSubspaceLDA(
-        priors=UnivariateFinite(classes(y), [1/3, 1/3, 1/3])
+        priors=UnivariateFinite(levels(y), [1/3, 1/3, 1/3])
     )
     fitresult2, cache2, report2 = fit(LDA_model2, 1, X, y)
     classes_seen2, projection_matrix2, priors2 = fitted_params(
@@ -231,24 +231,24 @@ end
     y2 = y[[1,2,1,2]]
     @test_throws ArgumentError fit(model, 1, X, y2)
 
-    ## Check to make sure error is thrown if UnivariateFinite `priors` doesn't have 
+    ## Check to make sure error is thrown if UnivariateFinite `priors` doesn't have
     ## common pool with target vector used in training.
     model = BayesianLDA(priors=UnivariateFinite([0.1, 0.5, 0.4], pool=missing))
     @test_throws ArgumentError fit(model, 1, X, y)
 
-    ## Check to make sure error is thrown if keys used in `priors` dictionary are in pool 
+    ## Check to make sure error is thrown if keys used in `priors` dictionary are in pool
     ## of training target used in training.
     model = BayesianLDA(priors=Dict("apples" => 0.1, "oranges"=>0.5, "bannana"=>0.4))
     @test_throws ArgumentError fit(model, 1, X, y)
 
     ## Check to make sure error is thrown if sum(`priors`) isn't approximately equal to 1.
-    model = BayesianLDA(priors=UnivariateFinite(classes(y), [0.1, 0.5, 0.4, 0.2]))
+    model = BayesianLDA(priors=UnivariateFinite(levels(y), [0.1, 0.5, 0.4, 0.2]))
     @test_throws ArgumentError fit(model, 1, X, y)
 
     ## Check to make sure error is thrown if `priors .< 0` or `priors .> 1`.
-    model = BayesianLDA(priors=Dict(classes(y) .=> [-0.1, 0.0, 1.0, 0.1]))
+    model = BayesianLDA(priors=Dict(levels(y) .=> [-0.1, 0.0, 1.0, 0.1]))
     @test_throws ArgumentError fit(model, 1, X, y)
-    model = BayesianLDA(priors=Dict(classes(y) .=> [1.1, 0.0, 0.0, -0.1]))
+    model = BayesianLDA(priors=Dict(levels(y) .=> [1.1, 0.0, 0.0, -0.1]))
     @test_throws ArgumentError fit(model, 1, X, y)
 
     X2 = (x=rand(5),)

test/runtests.jl

@@ -2,9 +2,11 @@ import Dates
 import MLJMultivariateStatsInterface: _replace!
 import MultivariateStats
 import Random
+import CategoricalDistributions.levels
 
 using LinearAlgebra
 using MLJBase
+using StatisticalMeasures
 using MLJMultivariateStatsInterface
 using StableRNGs
 using Test