add breaking test exposing issue #152

oops

oops again

test/core.jl

@@ -120,4 +120,41 @@ epochs = 10
 
 end
 
-true
+
+# to control chain initialization:
+myinit(n, m) = reshape(float(1:n*m), n , m)
+
+mutable struct TESTBuilder <: MLJFlux.Builder end
+MLJFlux.build(builder::TESTBuilder, n_in, n_out) =
+    Chain(Flux.Dense(n_in, n_out, init=myinit))
+
+@testset_accelerated "issue #152" accel begin
+
+    # data:
+    n = 100
+    d = 5
+    Xmat = rand(Float64, n, d)
+    X = MLJBase.table(Xmat);
+    y = X.x1 .^2 + X.x2 .* X.x3 - 4 * X.x4
+
+    # train a model on all the data using batch size > 1:
+    model = MLJFlux.NeuralNetworkRegressor(builder = TESTBuilder(),
+                                   batch_size=25,
+                                   epochs=1,
+                                   loss=Flux.mse)
+    mach = machine(model, X, y)
+    fit!(mach, verbosity=0)
+
+    # extract the pre-training loss computed in the `fit!(chain, ...)` method:
+    pretraining_loss = report(mach).training_losses[1]
+
+    # compute by hand:
+    chain0 = myinit(1, d)
+    pretraining_yhat = Xmat*chain0' |> vec
+    @test y isa Vector && pretraining_yhat isa Vector
+    pretraining_loss_by_hand =  MLJBase.l2(pretraining_yhat, y) |> mean
+
+    # compare:
+    @test_broken pretraining_loss ≈ pretraining_loss_by_hand
+
+end