Changeset 35 for trunk

Timestamp:

09/06/11 08:05:58 (9 months ago)

Author:

anonymous

Message:

Changing default configuration of the created network when the 'Combi' style and the 'selectBest' selection method is chosen. In that case, the trainParams.maxLayerNeurons has a value of 5. A large value for that parameter can lead to a very big structure during training, requiring great amount of memory. That is just a defualt value; it can be changed before train.

File:

• trunk/gmdh-t-ann/gmdhNew.m (modified) (4 diffs)

trunk/gmdh-t-ann/gmdhNew.m

@@ -2 +2 @@
 %
 % Creates a new GMDH network structure. With no more parameters than the
-% input count, the returned network will follow the default PNN structure,
-% where each layer has n!/(2(p-2)!) elements with two inputs, where n is
+% input count, the returned network will follow the default PNN structure:
+% each layer has n!/(2(p-2)!) elements with two inputs, where n is
 % the number of system inputs for the input layer or the number or elements
 % of the previous layer, for hidden layers.
@@ -26 +26 @@
 %                        contribute to the output of the network.
 %
-%    trainFunction       Specify the network training algorithm. Currently,
-%                        the available value is 'gmdhTrainMeanSquare'.
-%
 %    layerTrainFunction  Specify the layer training algorithm. Currently,
 %                        the available value is 'gmdhTrainMeanSquareLayer'.
@@ -37 +34 @@
 %                        to true.
 %    
-%   selectMethod         Defines the method for excluding the badly
+%    selectMethod        Defines the method for excluding the badly
 %                        performancing neurons in each layer during the 
 %                        training. Two methods are available:
@@ -100 +97 @@
     %
     % Basic train parameters
-    gmdhNet.trainParams.maxLayerNeurons = 30;
+    if gmdhNet.isCombi && strcmpi(selectMethod, 'selectBest')
+        gmdhNet.trainParams.maxLayerNeurons = 5;
+    else
+        gmdhNet.trainParams.maxLayerNeurons = 30;
+    end
     gmdhNet.trainParams.minLayerNeurons = 1;
     gmdhNet.trainParams.logs = true;