width first network

thorek1 · Oct 31, 2024 · e772b65 · e772b65
1 parent f5c9b13
commit e772b65
Showing 1 changed file with 14 additions and 12 deletions.
diff --git a/test/neural_net_solution.jl b/test/neural_net_solution.jl
@@ -197,12 +197,13 @@ outputs /= 6
 inputs /= 6
 # inputs .+= .5
 
-h5write("data.h5", "inputs", inputs)
-h5write("data.h5", "outputs", outputs)
+# h5write("data.h5", "inputs", inputs)
+# h5write("data.h5", "outputs", outputs)
 
 
 ## Create Neural Network
-n_hidden = max(256, n_vars * 2)
+n_hidden = max(1024, n_vars * 2)
+n_hidden_small = max(128, n_vars * 2)
 
 if recurrent
     neural_net = Chain( Dense(n_inputs, n_hidden, asinh),
@@ -215,12 +216,12 @@ if recurrent
 else
     if normalise
         neural_net = Chain( Dense(n_inputs, n_hidden, tanh_fast),
-                            Dense(n_hidden, n_hidden, leakyrelu),
-                            Dense(n_hidden, n_hidden, tanh_fast),
-                            Dense(n_hidden, n_hidden, leakyrelu),
-                            Dense(n_hidden, n_hidden, tanh_fast),
-                            Dense(n_hidden, n_hidden, leakyrelu),
-                            Dense(n_hidden, n_vars, tanh_fast))
+                            Dense(n_hidden, n_hidden_small, leakyrelu),
+                            Dense(n_hidden_small, n_hidden_small, tanh_fast),
+                            Dense(n_hidden_small, n_hidden_small, leakyrelu),
+                            Dense(n_hidden_small, n_hidden_small, tanh_fast),
+                            Dense(n_hidden_small, n_hidden_small, leakyrelu),
+                            Dense(n_hidden_small, n_vars, tanh_fast))
     else
         neural_net = Chain( Dense(n_inputs, n_hidden, asinh),
                             Dense(n_hidden, n_hidden, asinh),
@@ -254,8 +255,12 @@ if pretrain
     # BSON.@load "post_LBFGS.bson" neural_net
 end
 
+
+
 # Setup optimiser
 
+n_epochs = 300
+
 # optim = Flux.setup(Flux.Adam(), neural_net)
 optim = Flux.setup(Flux.Optimiser(Flux.ClipNorm(1), Flux.AdamW()), neural_net)
 
@@ -268,11 +273,8 @@ eta_sched = ParameterSchedulers.Stateful(CosAnneal(.001, 1e-10, n_epochs))
 #                                 [scheduler_period ÷ 3, scheduler_period ÷ 3, scheduler_period ÷ 3]))
 
 
-
 # Training loop
 
-n_epochs = 300
-
 batchsize = 1024
 
 train_loader = Flux.DataLoader((outputs, inputs), batchsize = batchsize, shuffle = true)