Merge pull request #260 from FluxML/optimiser-warning

Add check that Flux optimiser is not being used

src/mlj_model_interface.jl

@@ -7,6 +7,11 @@ MLJModelInterface.deep_properties(::Type{<:MLJFluxModel}) =
 
 # # CLEAN METHOD
 
+const ERR_BAD_OPTIMISER = ArgumentError(
+    "Flux.jl optimiser detected. Only optimisers from Optimisers.jl are supported. "*
+    "For example, use `optimiser=Optimisers.Momentum()` after `import Optimisers`. "
+)
+
 function MLJModelInterface.clean!(model::MLJFluxModel)
     warning = ""
     if model.lambda < 0
@@ -40,6 +45,9 @@ function MLJModelInterface.clean!(model::MLJFluxModel)
             "on an RNG during training, such as `Dropout`. Consider using "*
             " `Random.default_rng()` instead. `"
     end
+    # TODO: This could be removed in next breaking release (0.6.0):
+    model.optimiser isa Flux.Optimise.AbstractOptimiser && throw(ERR_BAD_OPTIMISER)
+
     return warning
 end
 
@@ -79,7 +87,7 @@ function regularized_optimiser(model, nbatches)
         Optimisers.WeightDecay(λ_weight),
         model.optimiser,
         )
-    end 
+    end
 end
 
 function MLJModelInterface.fit(model::MLJFluxModel,

test/mlj_model_interface.jl

@@ -35,6 +35,10 @@ end
         end
         @test model.acceleration == CUDALibs()
     end
+
+    @test_throws MLJFlux.ERR_BAD_OPTIMISER NeuralNetworkClassifier(
+        optimiser=Flux.Optimise.Adam(),
+    )
 end
 
 @testset "regularization: logic" begin