Why is the mean square value changing to an another value for a different training function ?

clc; clear all;close all;

seed = 2;

rand('seed',0);

co=1; % Testing takes pleace for co number of runs and average is returned

meanTrain=0; %mean performacne of training data

meanTest=0; %mean performacne of testing data

tn_data = xlsread('D:\PROJECT\back prop\train_data.xlsx');

s1 = tn_data;

tn_test = xlsread('D:\PROJECT\back prop\train_test.xlsx');

s2 = tn_test;

tt_data = xlsread('D:\PROJECT\back prop\test_data.xlsx');

s3 = tt_data;

tt_test = xlsread('D:\PROJECT\back prop\test_test.xlsx');

s4 = tt_test;

in=s(:,3:3); %in contains all inputs

ou=s(:,2:2); %ou contains all outputs

for i=1:co

    trainInput=[];

    trainOutput=[];

    testInput=[];

    testOutput=[];

    trainInput = s1;

    trainOutput = s2;

    testInput = s3;

    testOutput = s4;

    inn = [trainInput' testInput'];

    in = inn(:)

    % network initialization

    net=newff(minmax(in'),[20,1],{'tansig','purelin'},'traingdm_new');

    net.trainParam.show = 50;

    net.trainParam.lr = 0.01;

    net.trainParam.mc = 0.1;

    net.trainParam.epochs = 5000;

    net.trainParam.goal = 2e-2;

    % training

    [net,tr]=train(net,trainInput',trainOutput');

    % testing on training data

    a = sim(net,trainInput');

    a=a';

    corr=0;

    for j=1:size(trainInput,1)

        if a(j)< 0.49 ans=0; else ans=1; end

        if trainOutput(j)==ans corr=corr+1; 

        fprintf('\n the %d th image is correctly classified',j);

        end

    end

    corr; 

    corr/size(trainInput,1)*100; % percent train accuracy

    meanTrain=meanTrain+corr;

    % testing on testing data

    a = sim(net,testInput');

    a=a';

    corr=0;

    for j=1:size(testInput,1)

        if a(j)< 0.49 ans=0; else ans=1; end

        if testOutput(j)==ans corr=corr+1; 

            q(j) = corr;

        fprintf('\n the %d th image is correctly classified',j);

        end

    end

    corr;

    meanTest=meanTest+corr;

    corr/size(testInput,1)*100; % percent test accuracy

end

s=size(trainInput,1);

%'Mean Train Recognized'

ttt =meanTrain;

ttit=meanTrain/co;

'Mean Train Recognition Accuracy'

result1 = (meanTrain/co)/s*100 % mean percent train accuracy

s=size(testInput,1);

%'Mean Test Recognized'

ttet = meanTest;

tet = meanTest/co;

'Mean Test Recognition Accuracy'

result2 = (meanTest/co)/s*100 % mean percent test accuracy

e = testOutput - a;

perf1 = mse(e)

%%%%%%%%%%%%%%%%%%%%%%%%%%%% Code ends %%%%%%%%%%%%%%%%%%%%

The Test Accuracy that I get is 93.33 % and the mean square error is 0.1031

I have used the training function "traingdm".

If I use the training function "trainbr" and its corresponding training parameters, the classification percentage is 93.33%

But the mean square error changes to some other value. Why is this happening ????

[Jan]

Please format your code properly. This would increase the readability and therefore allows for creating an answer more efficiently. Thanks.

Answer 1

[Greg Heath]

 help trainbr

 Note that

 "trainbr ... minimizes a combination of squared errors and weights"

Hope this helps.

Greg