Merge pull request #209 from JuliaAI/dev

For a 0.8.2 release

Project.toml

@@ -1,7 +1,7 @@
 name = "MLJTuning"
 uuid = "03970b2e-30c4-11ea-3135-d1576263f10f"
 authors = ["Anthony D. Blaom <[email protected]>"]
-version = "0.8.1"
+version = "0.8.2"
 
 [deps]
 ComputationalResources = "ed09eef8-17a6-5b46-8889-db040fac31e3"

src/tuned_models.jl

@@ -1,4 +1,4 @@
-## TYPES AND CONSTRUCTOR
+# TYPES AND CONSTRUCTOR
 
 const ERR_SPECIFY_MODEL = ArgumentError(
 "You need to specify `model=...`, unless `tuning=Explicit()`. ")
@@ -687,7 +687,7 @@ function finalize(tuned_model,
                   history,
                   state,
                   verbosity,
-                  rm,
+                  resampling_machine,
                   data...)
     model = tuned_model.model
     tuning = tuned_model.tuning
@@ -713,7 +713,7 @@ function finalize(tuned_model,
     end
 
     report = merge(report1, tuning_report(tuning, history, state))
-    meta_state = (history, deepcopy(tuned_model), model_buffer, state, rm)
+    meta_state = (history, deepcopy(tuned_model), model_buffer, state, resampling_machine)
 
     return fitresult, meta_state, report
 end
@@ -749,9 +749,16 @@ function MLJBase.fit(tuned_model::EitherTunedModel{T,M},
     history, state = build!(nothing, n, tuning, model, model_buffer, state,
                            verbosity, acceleration, resampling_machine)
 
-    rm = resampling_machine
-    return finalize(tuned_model, model_buffer,
-                    history, state, verbosity, rm, data...)
+
+    return finalize(
+        tuned_model,
+        model_buffer,
+        history,
+        state,
+        verbosity,
+        resampling_machine,
+        data...,
+    )
 
 end
 
@@ -784,9 +791,15 @@ function MLJBase.update(tuned_model::EitherTunedModel,
         history, state = build!(history, n!, tuning, model, model_buffer, state,
                                verbosity, acceleration, resampling_machine)
 
-        rm = resampling_machine
-        return finalize(tuned_model, model_buffer,
-                        history, state, verbosity, rm, data...)
+        return finalize(
+            tuned_model,
+            model_buffer,
+            history,
+            state,
+            verbosity,
+            resampling_machine,
+            data...,
+        )
     else
         return  fit(tuned_model, verbosity, data...)
     end
@@ -806,6 +819,24 @@ function MLJBase.fitted_params(tuned_model::EitherTunedModel, fitresult)
 end
 
 
+## FORWARD SERIALIZATION METHODS FROM ATOMIC MODEL
+
+const ERR_SERIALIZATION = ErrorException(
+    "Attempting to serialize a `TunedModel` instance whose best model has not "*
+    "been trained. It appears as if it was trained with `train_best=false`. "*
+    "Try re-training using `train_best=true`. "
+)
+
+# `fitresult` is `machine(best_model, data...)`, trained iff `train_best` hyperparameter
+# is `true`.
+function MLJBase.save(tmodel::EitherTunedModel, fitresult)
+    MLJBase.age(fitresult) > 0 || throw(ERR_SERIALIZATION)
+    return MLJBase.serializable(fitresult)
+end
+MLJBase.restore(tmodel::EitherTunedModel, serializable_fitresult) =
+    MLJBase.restore!(serializable_fitresult)
+
+
 ## SUPPORT FOR MLJ ITERATION API
 
 MLJBase.iteration_parameter(::Type{<:EitherTunedModel}) = :n

test/tuned_models.jl

@@ -406,4 +406,53 @@ end
     MLJBase.Tables.istable(mach.cache[end].fitresult.machine.data[1])
 end
 
+# define a supervised model with ephemeral `fitresult`, but which overcomes this by
+# overloading `save`/`restore`:
+thing = []
+struct EphemeralRegressor <: Deterministic end
+function MLJBase.fit(::EphemeralRegressor, verbosity, X, y)
+    # if I serialize/deserialized `thing` then `id` below changes:
+    id = objectid(thing)
+    fitresult = (thing, id, mean(y))
+    return fitresult, nothing, NamedTuple()
+end
+function MLJBase.predict(::EphemeralRegressor, fitresult, X)
+    thing, id, μ = fitresult
+    return id == objectid(thing) ? fill(μ, nrows(X)) :
+        throw(ErrorException("dead fitresult"))
+end
+function MLJBase.save(::EphemeralRegressor, fitresult)
+    thing, _, μ = fitresult
+    return (thing, μ)
+end
+function MLJBase.restore(::EphemeralRegressor, serialized_fitresult)
+    thing, μ = serialized_fitresult
+    id = objectid(thing)
+    return (thing, id, μ)
+end
+
+@testset "save and restore" begin
+    # https://github.com/JuliaAI/MLJTuning.jl/issues/207
+    X, y = (; x = rand(10)), fill(42.0, 3)
+    tmodel = TunedModel(
+        models=fill(EphemeralRegressor(), 2),
+        measure=l2,
+        resampling=Holdout(),
+        train_best=false,
+    )
+    mach = machine(tmodel, X, y)
+    fit!(mach, verbosity=0)
+    io = IOBuffer()
+    @test_throws MLJTuning.ERR_SERIALIZATION MLJBase.save(io, mach)
+    close(io)
+    tmodel.train_best = true
+    fit!(mach, verbosity=0)
+    io = IOBuffer()
+    @test_logs MLJBase.save(io, mach)
+    seekstart(io)
+    mach2 = machine(io)
+    close(io)
+    @test MLJBase.predict(mach2, (; x = rand(2))) ≈ fill(42.0, 2)
+end
+
 true