Help for mathematical equation of regression in ANN

With ANN toolbox, I am using neural networks for finding the regression equation.

for info, I am using Bayesian regularization with 4 variables of 30 different samples and 30 results.

Is there a way of finding the mathematical equation of that in ANN?

Thanks..

Answer 1

[Greg Heath]

This question has been asked many times in both the Newsgroup and Answers. If you do not use the default normalizations of input and output,

h = tansig(IW*x+b1);

y = purelin(LW*h+b2);

Otherwise you have to use the default mapminmax or alternative mapstd on x,t and y.

You can obtain details by searching on the equation for h in the Newgroup and Answers.

Hope this helps.

Greg

[b]

Sorry for spamming.

I do not see these for equation.

Answer 2

[b]

Thank you so much but,

inputs = initial1';
      targets = output';
      hiddenLayerSize = 30;
      net = newff(minmax(input),[1 10],{'tansig' 'purelin'},'trainbr');
      net.IW{1}
      net.b{1,1}
      h=tansig(IW*inputs+b1)
      targets=purelin(LW*h+b2)  

    net.inputs{1}.processFcns = {'removeconstantrows','mapminmax'};
    net.outputs{2}.processFcns = {'removeconstantrows','mapminmax'};

    net.divideFcn = 'dividerand';  % Divide data randomly
    net.divideMode = 'sample';  % Divide up every sample
    net.divideParam.trainRatio = 80/100;
    net.divideParam.valRatio = 15/100;
    net.divideParam.testRatio = 5/100;

    net.trainFcn = 'trainbr';  % Bayesian Regularization

    net.performFcn = 'mse';  % Mean squared error

    net.plotFcns = {'plotperform','plottrainstate','ploterrhist', ...
      'plotregression', 'plotfit'};

    [net,tr] = train(net,inputs,targets);

    outputs = net(inputs);
    errors = gsubtract(targets,outputs);
    performance = perform(net,targets,outputs)

    trainTargets = targets .* tr.trainMask{1};
    valTargets = targets  .* tr.valMask{1};
    testTargets = targets  .* tr.testMask{1};
    trainPerformance = perform(net,trainTargets,outputs)
    valPerformance = perform(net,valTargets,outputs)
    testPerformance = perform(net,testTargets,outputs)

    view(net)

         perf = msereg(errors,outputs,X,FP);
         dy = msereg('dy',errors,outputs,X,perf,FP);
         dx = msereg('dx',errors,outputs,X,perf,FP);
         info = msereg(code);
         net.performParam.ratio = 20/(20+1);
         perf = msereg(e,net);

Gives an error why??

[Greg Heath]

What is the error?

Did you try to use the debugger?

Greg

[b]

I am using debugger, but actually,

net.IW{1}

net.b{1,1}

There is an error. I could not solve it.

Answer 3

[Greg Heath]

      hiddenLayerSize = 30;

      1. TOO LARGE AND INCOMPATIBLE WITH NEXT COMMAND

      net = newff(minmax(input),[1 10],{'tansig' 'purelin'},'trainbr');

     2. a. OBSOLETE. WHAT VERSION OF MATAB AND NNTBX DO YOU HAVE? 

     2.b.   INCORRECT NODE SIZE ASSIGNNMENT SYNTAX

      net.IW{1}
      net.b{1,1}

     3.  ASSIGN WEIGHTS TO IW, LW

      h=tansig(IW*inputs+b1)

      targets=purelin(LW*h+b2)  

      4. TERMINATE THIS AND OTHER VOLUMINOUS OUTPUT COMMANDS WITH SEMICOLONS

    net.inputs{1}.processFcns = {'removeconstantrows','mapminmax'};

    net.outputs{2}.processFcns = {'removeconstantrows','mapminmax'};

    net.divideFcn = 'dividerand';  % Divide data randomly

    net.divideMode = 'sample';  % Divide up every sample

    4. LAST FOUR ARE DEFAULTS: DELETE

     net.divideParam.valRatio = 15/100;

     5. WHY ARE YOU USING A VALIDATION SET WITH TRAINBR?

    net.trainFcn = 'trainbr';  % Bayesian Regularization   

    6. WHY ARE YOU USING TRAINBR INSTEAD OF DEFAULT TRAINLM?

    net.performFcn = 'mse';  % Mean squared error    

    7. MSE INCOMPATIBLE WITH TRAINBR SEE DOCUMENTATION

    net.plotFcns = {'plotperform','plottrainstate','ploterrhist', ...
      'plotregression', 'plotfit'};  

8. NOT SURE IF THESE ARE COMPATIBLE WITH YOUR OBSOLETE VERSION OF NEWFF

    perf = msereg(errors,outputs,X,FP);

         dy = msereg('dy',errors,outputs,X,perf,FP);

         dx = msereg('dx',errors,outputs,X,perf,FP);

         info = msereg(code);

         net.performParam.ratio = 20/(20+1);

         perf = msereg(e,net);

   9. I HAVE NO IDEA WHAT YOU ARE DOING HERE. YOU NEVER USED MSEREG FOR LEARNING

   HOPE THIS HELPS.

   GREG

Answer 4

[b]

hiddenLayerSize=10;
    # 1- HIDDENLAYERSIZE 30 CHANGED TO 10. 
    # 2- MATLAB 2012a.(FORGOT TO ADD IT, SORRY FOR THAT)
net.IW{1}
net.b{1,1}
    # 3- WHEN I WROTE getwb(net), IW AND B OCCUR LIKE THAT.
h=tansig(IW*inputs+b1);
targets=purelin(LW*h+b2);
    # 4- SEMICOLONS ARE ADDED.
net.trainFcn='trainlm'
    # 5- IT IS CHANGED TO 'TRAINLM'. IN MANY ARTICLES, BAYESIAN IS WORKED FOR MEDIA OPTIMIZATION. I CAN USE 'LM' IN MY PROJECT
net.plotFcns = {'plotperform','plottrainstate','ploterrhist', ...
    'plotregression', 'plotfit'};  
    # 6- THEY ARE COMPATIBLE AND WORKING WITHOUT AN ERROR.
dy = msereg('dy',errors,outputs,X,perf,FP);
dx = msereg('dx',errors,outputs,X,perf,FP);
info = msereg(code);
net.performParam.ratio = 20/(20+1);
perf = msereg(e,net);
    # 7- THESE FOR OBTAINIG MSEREG AND PERFORMANCE. 
CAN I USE 'ADAPT' MSEREG FOR LEARNING.

THANK YOU SO MUCH.

BASAR

[b]

The last sentence is not correct.

I misunderstood what you said in your last sentence.

[b]

What can i do in msereg for learning?

Answer 5

[b]

Also, in one of your previous answers, you told to use.

help msereg

in the help part, it is explained like that

perf = msereg(errors,outputs,X,FP);
         dy = msereg('dy',errors,outputs,X,perf,FP);
         dx = msereg('dx',errors,outputs,X,perf,FP);
         info = msereg(code);
         net.performParam.ratio = 20/(20+1);
         perf = msereg(e,net);