robustnn not as good as expected unfortunately

notebooks/mnist.qmd

@@ -475,16 +475,16 @@ function MLJFlux.build(b::LeNetBuilder, rng, n_in, n_out)
     preproc(x) = reshape(x, (_n_in, _n_in, 1, :))
 
 	front = Flux.Chain(
-        Conv((k, k), 1 => c1, pad=(p, p), relu),
+        Conv((k, k), 1 => c1, pad=(p, p), sigmoid),
         MaxPool((2, 2)),
-        Conv((k, k), c1 => c2, pad=(p, p), relu),
+        Conv((k, k), c1 => c2, pad=(p, p), sigmoid),
         MaxPool((2, 2)),
         Flux.flatten
     )
 	d = Flux.outputsize(front, (_n_in, _n_in, 1, 1)) |> first
     back = Flux.Chain(
-        Dense(d, 120, relu),
-        Dense(120, 84, relu),
+        Dense(d, 120, sigmoid),
+        Dense(120, 84, sigmoid),
         Dense(84, n_out),
     )
 
@@ -537,9 +537,13 @@ add_retrain = true
 # Deep Ensemble:
 mlp_large_ens = EnsembleModel(model=mlp, n=50)
 
+# CNN Ensemble:
+lenet_ens = EnsembleModel(model=lenet, n=5)
+
 add_models = Dict(
     "LeNet-5" => lenet,
-    "RobustNet" => rob_net,
+    "LeNet-5 Ensemble" => lenet_ens,
+    # "RobustNet" => rob_net,
     "Large Ensemble (n=50)" => mlp_large_ens,
 )
 
@@ -581,7 +585,8 @@ _plt_order = [
     "MLP Ensemble", 
     "Large Ensemble (n=50)", 
     "LeNet-5",
-    "RobustNet",
+    "LeNet-5 Ensemble",
+    # "RobustNet",
     "JEM", 
     "JEM Ensemble",
 ]
@@ -608,10 +613,9 @@ plts = [plt_additional_models]
 
 for (factual, target) in combos
     plt, _, _ = _plot_eccco_mnist(
+        factual, target; 
         plt_order = _plt_order,
         model_dict = large_model_dict,
-        x = factual,
-        target = target,
         wide = true,
         plot_factual = true,
         img_height = 150