Symstr (#45)

* penalties for basic regressions can be specified with  strings like sklearn

* patch release

Project.toml

@@ -1,7 +1,7 @@
 name = "MLJLinearModels"
 uuid = "6ee0df7b-362f-4a72-a706-9e79364fb692"
 authors = ["Thibaut Lienart <[email protected]>"]
-version = "0.2.3"
+version = "0.2.4"
 
 [deps]
 DocStringExtensions = "ffbed154-4ef7-542d-bbb7-c09d3a79fcae"

src/mlj/classifiers.jl

@@ -5,14 +5,14 @@
 @with_kw_noshow mutable struct LogisticClassifier <: MLJBase.Probabilistic
     lambda::Real             = 1.0
     gamma::Real              = 0.0
-    penalty::Symbol          = :l2
+    penalty::SymStr          = :l2
     fit_intercept::Bool      = true
     penalize_intercept::Bool = false
     solver::Option{Solver}   = nothing
     multi_class::Bool        = false
 end
 
-glr(m::LogisticClassifier) = LogisticRegression(m.lambda, m.gamma; penalty=m.penalty,
+glr(m::LogisticClassifier) = LogisticRegression(m.lambda, m.gamma; penalty=Symbol(m.penalty),
                                                 multi_class=m.multi_class,
                                                 fit_intercept=m.fit_intercept,
                                                 penalize_intercept=m.penalize_intercept)
@@ -26,13 +26,13 @@ descr(::Type{LogisticClassifier}) = "Classifier corresponding to the loss functi
 @with_kw_noshow mutable struct MultinomialClassifier <: MLJBase.Probabilistic
     lambda::Real             = 1.0
     gamma::Real              = 0.0
-    penalty::Symbol          = :l2
+    penalty::SymStr          = :l2
     fit_intercept::Bool      = true
     penalize_intercept::Bool = false
     solver::Option{Solver}   = nothing
 end
 
-glr(m::MultinomialClassifier) = MultinomialRegression(m.lambda, m.gamma; penalty=m.penalty,
+glr(m::MultinomialClassifier) = MultinomialRegression(m.lambda, m.gamma; penalty=Symbol(m.penalty),
                                                       fit_intercept=m.fit_intercept,
                                                       penalize_intercept=m.penalize_intercept)
 

src/mlj/interface.jl

@@ -2,6 +2,8 @@ export LinearRegressor, RidgeRegressor, LassoRegressor, ElasticNetRegressor,
        RobustRegressor, HuberRegressor, QuantileRegressor, LADRegressor,
        LogisticClassifier, MultinomialClassifier
 
+const SymStr = Union{Symbol,String}
+
 include("regressors.jl")
 include("classifiers.jl")
 

src/mlj/regressors.jl

@@ -69,13 +69,13 @@ descr(::Type{ElasticNetRegressor}) = "Regression with objective function ``|Xθ
     rho::RobustRho           = HuberRho(0.1)
     lambda::Real             = 1.0
     gamma::Real              = 0.0
-    penalty::Symbol          = :l2
+    penalty::SymStr          = :l2
     fit_intercept::Bool      = true
     penalize_intercept::Bool = false
     solver::Option{Solver}   = nothing
 end
 
-glr(m::RobustRegressor) = RobustRegression(m.rho, m.lambda, m.gamma; penalty=m.penalty,
+glr(m::RobustRegressor) = RobustRegression(m.rho, m.lambda, m.gamma; penalty=Symbol(m.penalty),
                                            fit_intercept=m.fit_intercept,
                                            penalize_intercept=m.penalize_intercept)
 
@@ -89,13 +89,13 @@ descr(::Type{RobustRegressor}) = "Robust regression with objective ``∑ρ(Xθ -
     delta::Real              = 0.5
     lambda::Real             = 1.0
     gamma::Real              = 0.0
-    penalty::Symbol          = :l2
+    penalty::SymStr          = :l2
     fit_intercept::Bool      = true
     penalize_intercept::Bool = false
     solver::Option{Solver}   = nothing
 end
 
-glr(m::HuberRegressor) = HuberRegression(m.delta, m.lambda, m.gamma; penalty=m.penalty,
+glr(m::HuberRegressor) = HuberRegression(m.delta, m.lambda, m.gamma; penalty=Symbol(m.penalty),
                                          fit_intercept=m.fit_intercept,
                                          penalize_intercept=m.penalize_intercept)
 
@@ -109,13 +109,14 @@ descr(::Type{HuberRegressor}) = "Robust regression with objective ``∑ρ(Xθ -
     delta::Real              = 0.5
     lambda::Real             = 1.0
     gamma::Real              = 0.0
-    penalty::Symbol          = :l2
+    penalty::SymStr          = :l2
     fit_intercept::Bool      = true
     penalize_intercept::Bool = false
     solver::Option{Solver}   = nothing
 end
 
-glr(m::QuantileRegressor) = QuantileRegression(m.delta, m.lambda, m.gamma; penalty=m.penalty,
+glr(m::QuantileRegressor) = QuantileRegression(m.delta, m.lambda, m.gamma;
+                                               penalty=Symbol(m.penalty),
                                                fit_intercept=m.fit_intercept,
                                                penalize_intercept=m.penalize_intercept)
 
@@ -128,13 +129,13 @@ descr(::Type{QuantileRegressor}) = "Robust regression with objective ``∑ρ(Xθ
 @with_kw_noshow mutable struct LADRegressor <: MLJBase.Deterministic
     lambda::Real             = 1.0
     gamma::Real              = 0.0
-    penalty::Symbol          = :l2
+    penalty::SymStr          = :l2
     fit_intercept::Bool      = true
     penalize_intercept::Bool = false
     solver::Option{Solver}   = nothing
 end
 
-glr(m::LADRegressor) = LADRegression(m.lambda, m.gamma; penalty=m.penalty,
+glr(m::LADRegressor) = LADRegression(m.lambda, m.gamma; penalty=Symbol(m.penalty),
                                      fit_intercept=m.fit_intercept,
                                      penalize_intercept=m.penalize_intercept)
 

test/interface/fitpredict.jl

@@ -60,3 +60,12 @@ end
     mcr = MLJBase.misclassification_rate(ŷ, yc)
     @test mcr ≤ 0.2
 end
+
+# see issue https://github.com/alan-turing-institute/MLJ.jl/issues/387
+@testset "String-Symbol" begin
+    model = LogisticClassifier(penalty="l1")
+    @test model.penalty == "l1"
+    gr = MLJLinearModels.glr(model)
+    @test gr isa GLR
+    @test gr.penalty isa ScaledPenalty{L1Penalty}
+end