For a 0.5.1 release (#156)

* add `feature_importances` stub  (#148)

* add intrinsic_importances stub and set fallback to
othing.

* fix error in intrinsic_importances docstring.

* rename intrinsic_importances method to eature_importances.

* remove fallback for eature_importances.

* Update src/model_api.jl

Co-authored-by: Anthony Blaom, PhD <[email protected]>

* bump 1.5

* Update `metadata_model` to include traits for feature importances and training losses (#155)

* + supports_training_losses,reports_feature_importances to model_metadata

* bump 1.5.1

* bump StatisticalTraits = "3.1"

Co-authored-by: Okon Samuel <[email protected]>

Project.toml

@@ -1,7 +1,7 @@
 name = "MLJModelInterface"
 uuid = "e80e1ace-859a-464e-9ed9-23947d8ae3ea"
 authors = ["Thibaut Lienart and Anthony Blaom"]
-version = "1.5"
+version = "1.5.1"
 
 [deps]
 Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
@@ -10,7 +10,7 @@ StatisticalTraits = "64bff920-2084-43da-a3e6-9bb72801c0c9"
 
 [compat]
 ScientificTypesBase = "3.0"
-StatisticalTraits = "3.0"
+StatisticalTraits = "3.1"
 julia = "1"
 
 [extras]

src/metadata_utils.jl

@@ -85,6 +85,9 @@ Helper function to write the metadata for a model `T`.
 * `supports_class_weights=false`: whether the model supports class weights
 * `load_path="unknown"`: where the model is (usually `PackageName.ModelName`)
 * `human_name=nothing`: human name of the model
+* `supports_training_losses=nothing`: whether the (necessarily iterative) model can report
+  training losses
+* `reports_feature_importances=nothing`: whether the model reports feature importances
 
 ## Example
 
@@ -115,7 +118,9 @@ function metadata_model(
     supports_class_weights::Union{Nothing,Bool}=class_weights,
     docstring::Union{Nothing,String}=descr,
     load_path::Union{Nothing,String}=path,
-    human_name::Union{Nothing,String}=nothing
+    human_name::Union{Nothing,String}=nothing,
+    supports_training_losses::Union{Nothing,Bool}=nothing,
+    reports_feature_importances::Union{Nothing,Bool}=nothing,
 )
     docstring === nothing || Base.depwarn(DEPWARN_DOCSTRING, :metadata_model)
 
@@ -132,6 +137,8 @@ function metadata_model(
     _extend!(program, :docstring, docstring, T)
     _extend!(program, :load_path, load_path, T)
     _extend!(program, :human_name, human_name, T)
+    _extend!(program, :supports_training_losses, supports_training_losses, T)
+    _extend!(program, :reports_feature_importances, reports_feature_importances, T)
 
     parentmodule(T).eval(program)
 end

src/model_api.jl

@@ -41,7 +41,7 @@ in historical order. If the model calculates scores instead, then the
 sign of the scores should be reversed.
 
 The following trait overload is also required:
-`supports_training_losses(::Type{<:M}) = true`
+`MLJModelInterface.supports_training_losses(::Type{<:M}) = true`.
 
 """
 training_losses(model, report) = nothing
@@ -168,16 +168,17 @@ function evaluate end
 """
     feature_importances(model::M, fitresult, report)
 
-For a given `model` of model type `M` supporting intrinsic feature importances, calculate 
-the feature importances from the model's `fitresult` and `report` as an 
-abstract vector of `feature::Symbol => importance::Real` pairs 
-(e.g `[:gender =>0.23, :height =>0.7, :weight => 0.1]`).  
+For a given `model` of model type `M` supporting intrinsic feature importances, calculate
+the feature importances from the model's `fitresult` and `report` as an
+abstract vector of `feature::Symbol => importance::Real` pairs
+(e.g `[:gender =>0.23, :height =>0.7, :weight => 0.1]`).
 
 The following trait overload is also required:
-`reports_feature_importances(::Type{<:M}) = true`
+`MLJModelInterface.reports_feature_importances(::Type{<:M}) = true`
 
 If for some reason a model is sometimes unable to report feature importances then
-`feature_importances` should return all importances as 0.0, as in 
-`[:gender =>0.0, :height =>0.0, :weight => 0.0]`. 
+`feature_importances` should return all importances as 0.0, as in
+`[:gender =>0.0, :height =>0.0, :weight => 0.0]`.
+
 """
 function feature_importances end

test/metadata_utils.jl

@@ -142,6 +142,8 @@ infos =  Dict(trait => eval(:(MLJModelInterface.$trait))(FooRegressor) for
     @test infos[:hyperparameters] == (:a, :b)
     @test infos[:hyperparameter_types] == ("Int64", "Any")
     @test infos[:hyperparameter_ranges] == (nothing, nothing)
+    @test !infos[:supports_training_losses] 
+    @test !infos[:reports_feature_importances]
 end
 
 @testset "doc_header(ModelType)" begin

test/model_api.jl

@@ -14,12 +14,22 @@ mutable struct APIx1 <: Static end
     @test selectrows(APIx0(), 2:3, X, y) == ((x1 = [4, 6],), [20.0, 30.0])
 end
 
+M.metadata_model(
+    APIx0,
+    supports_training_losses = true,
+    reports_feature_importances = true,
+)
+
+dummy_losses = [1.0, 2.0, 3.0]
+M.training_losses(::APIx0, report) = report
+M.feature_importances(::APIx0, fitresult, report) = [:a=>0, :b=>0]
+
 @testset "fit-x" begin
     m0 = APIx0(f0=1)
     m1 = APIx0b(f0=3)
     # no weight support: fallback
-    M.fit(m::APIx0, v::Int, X, y) = (5, nothing, nothing)
-    @test fit(m0, 1, randn(2), randn(2), 5) == (5, nothing, nothing)
+    M.fit(m::APIx0, v::Int, X, y) = (5, nothing, dummy_losses)
+    @test fit(m0, 1, randn(2), randn(2), 5) == (5, nothing, dummy_losses)
     # with weight support: use
     M.fit(m::APIx0b, v::Int, X, y, w) = (7, nothing, nothing)
     @test fit(m1, 1, randn(2), randn(2), 5) == (7, nothing, nothing)
@@ -32,16 +42,18 @@ end
     @test fit(s1, 1, 0) == (nothing, nothing, nothing)
 
     # update fallback = fit
-    @test update(m0, 1, 5, nothing, randn(2), 5) == (5, nothing, nothing)
+    @test update(m0, 1, 5, nothing, randn(2), 5) == (5, nothing, dummy_losses)
 
     # training losses:
     f, c, r = MLJModelInterface.fit(m0, 1, rand(2), rand(2))
-    @test M.training_losses(m0, r) === nothing
-
-    # intrinsic_importances
+    @test M.training_losses(m0, r) == dummy_losses
+
+    # training losses:
+    f, c, r = MLJModelInterface.fit(m0, 1, rand(2), rand(2))
+    @test M.training_losses(m0, r) == dummy_losses
+
+    # feature_importances
     f, c, r = MLJModelInterface.fit(m0, 1, rand(2), rand(2))
-    MLJModelInterface.reports_feature_importances(::Type{APIx0}) = true
-    MLJModelInterface.feature_importances(::APIx0, fitresult, report) = [:a=>0, :b=>0]
     @test MLJModelInterface.feature_importances(m0, f, r) == [:a=>0, :b=>0]
 end
 
@@ -67,7 +79,7 @@ mutable struct UnivariateFiniteFitter <: Probabilistic end
     end
 
     MMI.input_scitype(::Type{<:UnivariateFiniteFitter}) = Nothing
-    
+
     MMI.target_scitype(::Type{<:UnivariateFiniteFitter}) = AbstractVector{<:Finite}
 
     y = categorical(collect("aabbccaa"))