Merge pull request #15 from JuliaAI/dev

For a 0.2.2 release

Project.toml

@@ -1,7 +1,7 @@
 name = "MLJIteration"
 uuid = "614be32b-d00c-4edb-bd02-1eb411ab5e55"
 authors = ["Anthony D. Blaom <[email protected]>"]
-version = "0.2.1"
+version = "0.2.2"
 
 [deps]
 IterationControl = "b3c1a2ee-3fec-4384-bf48-272ea71de57c"
@@ -14,11 +14,10 @@ MLJBase = "0.17.7, 0.18"
 julia = "1"
 
 [extras]
-CategoricalArrays = "324d7699-5711-5eae-9e2f-1d82baa6b597"
 MLJModelInterface = "e80e1ace-859a-464e-9ed9-23947d8ae3ea"
 StableRNGs = "860ef19b-820b-49d6-a774-d7a799459cd3"
 Statistics = "10745b16-79ce-11e8-11f9-7d13ad32a3b2"
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 
 [targets]
-test = ["CategoricalArrays", "MLJModelInterface", "StableRNGs", "Statistics", "Test"]
+test = ["MLJModelInterface", "StableRNGs", "Statistics", "Test"]

src/MLJIteration.jl

@@ -12,6 +12,8 @@ const CONTROLS = vcat(IterationControl.CONTROLS,
                        :WithEvaluationDo,
                        :CycleLearningRate])
 
+const TRAINING_CONTROLS = [:Step, ]
+
 # export all control types:
 for control in CONTROLS
     eval(:(export $control))
@@ -24,11 +26,16 @@ const CONTROLS_DEFAULT = [Step(10),
                           NotANumber()]
 
 include("utilities.jl")
+include("controls.jl")
+
+const Control = Union{[@eval($c) for c in CONTROLS]...}
+const TrainingControl = Union{[@eval($c) for c in TRAINING_CONTROLS]...}
+
 include("constructors.jl")
 include("traits.jl")
 include("ic_model.jl")
-include("controls.jl")
 include("core.jl")
 
 
+
 end # module

src/constructors.jl

@@ -1,3 +1,9 @@
+const ERR_MISSING_TRAINING_CONTROL =
+    ArgumentError("At least one control must be a training control "*
+                  "(have type `$TrainingControl`) or be a "*
+                  "custom control that calls IterationControl.train!. ")
+
+
 ## TYPES AND CONSTRUCTOR
 
 mutable struct DeterministicIteratedModel{M<:Deterministic} <: MLJBase.Deterministic
@@ -176,8 +182,6 @@ function IteratedModel(; model=nothing,
 
     model == nothing && throw(ERR_NO_MODEL)
 
-
-
     if model isa Deterministic
         iterated_model = DeterministicIteratedModel(model,
                                                     controls,
@@ -229,5 +233,20 @@ function MLJBase.clean!(iterated_model::EitherIteratedModel)
         iteration_parameter(iterated_model.model) === nothing &&
         throw(ERR_NEED_PARAMETER)
 
+    if iterated_model.resampling isa Holdout &&
+        iterated_model.resampling.shuffle
+        message *= "The use of sample-shuffling in `Holdout` "*
+            "will significantly slow training as "*
+            "each increment of the iteration parameter "*
+            "will force iteration from scratch (cold restart). "
+    end
+
+    training_control_candidates = filter(iterated_model.controls) do c
+        c isa TrainingControl || !(c isa Control)
+    end
+    if isempty(training_control_candidates)
+        throw(ERR_MISSING_TRAINING_CONTROL)
+    end
+
     return message
 end

test/_dummy_model.jl

@@ -5,7 +5,7 @@ export DummyIterativeModel, make_dummy
 using Random
 using Statistics
 import StableRNGs.LehmerRNG
-using CategoricalArrays
+using MLJBase.CategoricalArrays
 import Base.==
 
 using MLJModelInterface

test/constructors.jl

@@ -21,7 +21,17 @@ struct Bar <: MLJBase.Deterministic end
     @test iterated_model.measure == RootMeanSquaredError()
     @test_logs IteratedModel(model=model, measure=mae)
 
-    iterated_model = @test_logs IteratedModel(model=model, resampling=nothing)
+    @test_logs IteratedModel(model=model, resampling=nothing)
+
+    @test_logs((:info, r"The use of sample"),
+               IteratedModel(model=model,
+                             resampling=Holdout(rng=123),
+                             measure=rms))
+
+    @test_throws(MLJIteration.ERR_MISSING_TRAINING_CONTROL,
+                 IteratedModel(model=model,
+                               resampling=nothing,
+                               controls=[Patience(), NotANumber()]))
 end
 
 end