First commit

Assignment 1/Car Evaluation/Neural.m

@@ -0,0 +1,55 @@
+function Neural(fold,n,classes)
+%% This function use x-fold cross validation to explore the performance of 
+% Neural Network with different hyper parameters. 
+
+% fold - the number of fold to perform cross validation
+% n - the number of neurons
+% the number of classes present in the dataset
+
+load('training_test.mat');
+
+indices = crossvalind('Kfold',training_label,fold);
+func = {'trainlm','trainbfg','trainrp','trainscg','traincgb','traincgf','traincgp','trainoss','traingdx'};
+rate = zeros(length(func),fold);
+hiddenLayerSize = 1:10;
+
+for k = 1:length(func)
+    disp(['Training function: ' func{k}]);
+    for j = 1:length(hiddenLayerSize)
+        disp(['hidden layer size: ' num2str(j)]);
+        nnp = zeros(1,10);
+        for i = 1:fold
+            valid = (indices == i); training = ~valid;
+            valid_d = training_data(valid,:);
+            valid_l = training_label(valid);
+
+            train_d = training_data(training,:);
+            train_l = training_label(training);
+
+            target = zeros(classes,length(train_l));
+            for x = 1:classes
+                target(x,:) = train_l==x;
+            end
+
+            neurons = ones(1,i) * n;
+
+            net = fitnet(neurons);
+            net.trainFcn = func{k};
+            net.trainParam.showWindow = false;
+
+            %     Train the Network
+            net = train(net,train_d',target);
+
+            % Test the Network
+            result = net(valid_d');
+            nnResult = vec2ind(result);
+            nnp(i) = sum(nnResult' == valid_l)/length(nnResult);
+        end
+        rate(k,j) = mean(nnp);
+    end
+
+end
+save(['neural_result_' num2str(n) '.mat']);
+end
+
+