Merge branch 'dev'

Project.toml

@@ -1,7 +1,7 @@
 name = "MLJLinearModels"
 uuid = "6ee0df7b-362f-4a72-a706-9e79364fb692"
 authors = ["Thibaut Lienart <tlienart@me.com>"]
-version = "0.7.0"
+version = "0.7.1"
 
 [deps]
 DocStringExtensions = "ffbed154-4ef7-542d-bbb7-c09d3a79fcae"
@@ -25,6 +25,7 @@ julia = "1.5"
 DataFrames = "a93c6f00-e57d-5684-b7b6-d8193f3e46c0"
 DelimitedFiles = "8bb1440f-4735-579b-a4ab-409b98df4dab"
 ForwardDiff = "f6369f11-7733-5829-9624-2563aa707210"
+MLJ = "add582a8-e3ab-11e8-2d5e-e98b27df1bc7"
 MLJBase = "a7f614a8-145f-11e9-1d2a-a57a1082229d"
 PyCall = "438e738f-606a-5dbb-bf0a-cddfbfd45ab0"
 RCall = "6f49c342-dc21-5d91-9882-a32aef131414"
@@ -34,4 +35,4 @@ StableRNGs = "860ef19b-820b-49d6-a774-d7a799459cd3"
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 
 [targets]
-test = ["DelimitedFiles", "PyCall", "ForwardDiff", "Test", "Random", "RDatasets", "RCall", "MLJBase", "StableRNGs", "DataFrames"]
+test = ["DelimitedFiles", "PyCall", "ForwardDiff", "Test", "Random", "RDatasets", "RCall", "MLJ", "MLJBase", "StableRNGs", "DataFrames"]

src/mlj/interface.jl

@@ -62,7 +62,11 @@ function MMI.fit(m::Union{CLF_MODELS...}, verb::Int, X, y)
     classes  = MMI.classes(y[1])
     nclasses = length(classes)
     if nclasses < 2
-        throw(DomainError("The target `y` needs to have two or more levels."))
+        throw(
+            DomainError(
+                "The target `y` needs to have two or more levels."
+            )
+        )
     elseif nclasses == 2 && m isa LogisticClassifier
         # recode to ± 1
         yplain[yplain .== 1] .= -1
@@ -71,8 +75,11 @@ function MMI.fit(m::Union{CLF_MODELS...}, verb::Int, X, y)
         # in the Logistic case and not in the Multinomial {0} case!
         nclasses = 0
     end
-    # NOTE: here the number of classes is either 0 or > 2
-    clf = glr(m, nclasses)
+    # NOTE: here nclasses is either
+    #   - 0  Logistic, binary
+    #   - 2  Multinomial, binary
+    #   - >2 or > 2 (Multinomial)
+    clf    = glr(m, nclasses)
     solver = m.solver === nothing ? _solver(clf, size(Xmatrix)) : m.solver
     verb > 0 && @info "Solver: $(solver)"
     # get the parameters
@@ -84,13 +91,16 @@ end
 function MMI.predict(m::Union{CLF_MODELS...}, (θ, features, classes, c), Xnew)
     Xmatrix = MMI.matrix(Xnew)
     preds = apply_X(Xmatrix, θ, c)
-    if c > 2 # multiclass
+
+    if c > 2 || m isa MultinomialClassifier # multiclass
         preds .= softmax(preds)
-    else # binary (necessarily c==0)
+
+    else # binary logistic (necessarily c==0)
         preds  .= sigmoid.(preds)
         preds   = hcat(1.0 .- preds, preds) # scores for -1 and 1
         return MMI.UnivariateFinite(classes, preds)
     end
+
     return MMI.UnivariateFinite(classes, preds)
 end
 

test/interface/extras.jl

@@ -24,3 +24,22 @@ end
     y = MLJBase.coerce(vcat(fill("a", 10), ["b", ]), MLJBase.Multiclass)[1:10]
     mach = MLJBase.machine(MultinomialClassifier(), X, y) |> MLJBase.fit!
 end
+
+# https://github.com/JuliaAI/MLJLinearModels.jl/issues/129
+@testset "Crabs" begin
+    data = MLJ.load_crabs()
+    y_, X = MLJ.unpack(data, ==(:sp), col->col in [:FL, :RW]);
+    y = MLJ.coerce(y_, MLJ.OrderedFactor);
+    model = MultinomialClassifier()
+    mach = MLJ.machine(model, X, y) |> MLJ.fit!
+    yhat = MLJ.predict_mode(mach, X)
+
+    # crappy "test" but we're just testing that predict works fine    
+    @test MLJ.misclassification_rate(yhat, y) < 0.4
+
+    model = LogisticClassifier()
+    mach = MLJ.machine(model, X, y) |> MLJ.fit!
+    yhat = MLJ.predict_mode(mach, X)
+
+    @test MLJ.misclassification_rate(yhat, y) < 0.4
+end

test/runtests.jl

@@ -1,6 +1,6 @@
 using MLJLinearModels, Test, LinearAlgebra
 using Random, StableRNGs, DataFrames, ForwardDiff
-import MLJBase # not MLJModelInterface, to mimic the full interface
+import MLJ, MLJBase
 
 DO_COMPARISONS = false; include("testutils.jl")