terrains + more than 2 layers

matifraga · May 7, 2017 · 47ed369 · 47ed369
1 parent 0d3a77d
commit 47ed369
Show file tree

Hide file tree

Showing 12 changed files with 4,884 additions and 8 deletions.
diff --git a/multilayer_perceptron.m b/multilayer_perceptron.m
@@ -1,11 +1,25 @@
 2;
-function output = get_output(entry,weights,activation_func)
-    output = activation_func(weights * entry);
+function output = get_output(entries, weights, neurons_per_layer, activation_func)
+    m = 0;
+    for i = 2:length(neurons_per_layer)
+        m = m + 1;
+        layer_entry{m} = [-1, zeros(1, neurons_per_layer(i-1))];
+    end
+    M = m;
+    i = 1;
+    for entry = entries
+        layer_entry{1}(2:end) = entry;
+        for m = 2:M
+            layer_entry{m}(2:end) = activation_func(weights{m-1} * layer_entry{m-1}');
+        end
+        output(i) = activation_func(weights{M} * layer_entry{M}');
+        i = i + 1;
+    end   
 end
 
 function [weights,output,error] = multilayer_perceptron_learn(entries, expected_output, neurons_per_layer, activation_func, activation_der,
                                     learning_factor=.5, max_iterations=1000, tolerance=1e-5)
-    n = length(entries(:, 1));
+    n = length(entries(1,:));
     m = 0;
     for i = 2:length(neurons_per_layer)
         m = m + 1;
@@ -14,8 +28,8 @@
     end
     M = m;
     for iteration = 1:max_iterations
-        for index = randperm(2^n); 
-                layer_entry{1}(2:end) = entries(:, index);
+        for index = randperm(n); 
+            layer_entry{1}(2:end) = entries(:, index);
             for m = 2:M
                 layer_entry{m}(2:end) = activation_func(weights{m-1} * layer_entry{m-1}');
             end
@@ -24,13 +38,13 @@
             output(index);
             d{M} = activation_der(weights{M}*layer_entry{M}')*(expected_output(index) - output(index));
             for i = M-1:-1:1
-                d{i} = (weights{i+1}(:,2:end)*d{i+1}' * activation_der(weights{i}*layer_entry{i}'));
+                d{i} = (activation_der(weights{i}*layer_entry{i}')' .* sum(weights{i+1}(:,2:end) .* d{i+1}'))';
             end
             d;
             weights;
             for i = 1:M
-                delta_w = learning_factor * layer_entry{i}'*d{i};
-                weights{i} = weights{i} + delta_w';
+                delta_w = learning_factor * d{i} * layer_entry{i};
+                weights{i} = weights{i} + delta_w;
             end
         end
         error(iteration) = sum(abs(expected_output - output));