Help meeeee.....through this prblm.

Question

0 个投票

hai, i want a program on three input and two output neural network with two hidden layers(1st hidden layer should have 3 nodes and 2nd layer 2 nodes... )

this program help me in solving so many things....so ,pls send the prgm as early as possible...

thank you,

2 个评论
显示无隐藏无

Greg Heath 2012-2-14

Usually there is no good reason to use more than one hidden layer. Similarly, there is usually no good reason to not use as many default parameters as possible.

What version of the NNtoolbox are you using? There are currently two versions of NEWFF, and one each of FEEDFORWARDNET and FITNET.

See samples of code in the Documentation, the Newsgroup, and Answers.

Hope this helps.

Greg

Greg Heath 2012-2-14

What type of net do you need?

Regression (curve fitting) , Classification, or Time-Series Prediction?

Greg

请先登录，再进行评论。

请先登录，再回答此问题。

Follow Question

Answer 1

Greg Heath 2012-2-14

0 个投票

The syntax to create a feedforward network with :

- 3 unit of inputs - hidden1 : 3 nodes - hidden2 : 2 nodes - 2 unit of outputs is shown below :

net = newff(PR,[3 3 2 2],{'logsig','logsig','logsig','logsig'},'traingda');

INCORRECT. THIS SPECIFIES 3 HIDDEN LAYERS. ONLY 1 IS NEEDED.

NOTE: I AM USING ALL CAPS FOR CLARIFICATION. DO NOT USE ALL CAPS IN YOUR CODE!

EITHER STANDARDIZE INPUTS TO ZERO-MEAN/UNIT-VARIANCE OR, IF PHYSICAL OR MATHEMATICAL BOUNDS ARE KNOWN, NORMALIZE TO THE BOUNDED BIPOLAR INTERVAL [-1,1].

FOR REGRESSION STANDARDIZE OUTPUTS.

USE 1 HIDDEN LAYER

USE TANSIG FOR HIDDEN NODES (DEFAULT)

USE PURELIN FOR THE OUTPUT NODES (DEFAULT)

USE TRAINLM FOR TRAINING (DEFAULT)

IF Pn IS THE NORMALIZED INPUT MATRIX AND Tn IS THE NORMALIZED OUTPUT MATRIX,

[ I N ] = SIZE(Pn)

[O N ] = SIZE( Tn)

FOR I-H-O NODE TOPOLOGY

NET = NEWFF(MINMAX(Pn), [H O]); % OLD VERSION

NET = NEWFF(Pn,Tn,H); % NEWER VERSION

NET = FITNET(H); % NEWEST VERSION

SEE THE DOCUMENTATION EXAMPLES ...(E.G., DOC FITNET)

In the next step you need to train this network. You can use train command from the toolbox.

For stopping criteria you can set the maximum number of epochs : 1500 and performance goal : 0

net.trainParam.epochs = 1500;

net.trainParam.goal = 0;

IF YOU USE TRAINLM THE DEFAULT IS 100 EPOCHS. INCREASE IF NEEDED

A PRACTICAL GOAL IS MEAN(VAR(Tn'))/100

net = train(net,pa,t);

Then to simulate the output, you can use sim command.

[NET TR Y] = TRAIN(...) AUTOMATICALLY YIELDS THE OUTPUT Y

This code below is just an example for the complete code :

p = [4 3 4 1 3 2]; t = eye(2);

COLUMNS OF EYE TYPICALLY ONLY USED FOR CLASSIFICATION.

pa = (p-min(p(:))) / (max(p(:))-min(p(:)));

PR = [0 1;0 1;0 1];

NOT RECOMMENDED. NORMALIZE EACH ROW SEPARATELY. EITHER STANDARDIZE OR NORMALIZE TO [-1 ,1]

-----SNIP

GREG

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

uday KUMAR 2012-2-23

Sir,I need matlab code.....this code also helps me but....for me classification of outputs is needed.Below this i am sending the code which i had written,,,and rectify the classification pblm...

thank u

请先登录，再进行评论。

Answer 2

uday KUMAR 2012-2-23

在 MATLAB Online 中打开

0 个投票

Hello sir, this is the code i had implemented for cancer dataset,her i am using 2 hiddenlayers,,,,,,but here i am not able to classify(classification of outputs).So ,pls solve this error and if modification is needed modify it,and let me know what the error is,,,,,,

        I am so eagrly waiting for u r reply.Pls send the suitable suggestion as early as possible.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%...............
tpat = xlsread('cancer.xls');
tpat(:,2:end) = zscore(tpat(:,2:end));
ni=length(tpat(1,:))-1;%no of inputs
no=2;%no of output
ntmax=length(tpat(:,1));%no of patterns
nh=4;%no of neurons in  1st hidden layer
nl=3;%no of neurons in  2nd hidden layer
itmax=500;%maximum iterations
eta=5;%learning rate 
wh=rand(nh,ni);
wl=rand(nl,nh)-0.5;
wo=rand(2,nl)-0.5;
bh=rand(1,nh)-0.5;
bl=rand(1,nl)-0.5;
bo=rand(1,2)-0.5;
%training routine.......... 
for it = 1:itmax
    for nt = 1:ntmax
        x = tpat(nt,2:ni+1);
        t = tpat(nt,1);
        neth = (wh*x')'+bh;
        h=1./(1+exp(-neth));
        neth1 = (wl*h')'+bl;
        k=1./(1+exp(-neth1));
        nety = (wo*k')'+bo;
        y=1./(1+exp(-nety));
    %Backward Propagation Starts..............
        e=eta*(t-y).*y.*(1-y);
        dbo=e;
        dwo=e'*k;
        dbl=(e*wo).*k.*(1-k);
        dwl=dbl'*h;
        dbh=(dbl*wl).*h.*(1-h);
        dwh=dbh'*x;
        bo=bo+dbo;
        bl=bl+dbl;
        bh=bh+dbh;
        wo=wo+dwo;
        wl=wl+dwl;
        wh=wh+dwh;
    end
    for nt = 1:ntmax
        x = tpat(nt,2:ni+1);
        t = tpat(nt,1);
        neth = (wh*x')'+bh;
        h=1./(1+exp(-neth));
        neth1 = (wl*h')'+bl;
        k=1./(1+exp(-neth1));
        nety = (wo*k')'+bo;
        y=1./(1+exp(-nety));
        ee(nt,:)=t-y;
    end
      eee(it,:)=sum(ee(nt,:).^2)/2/ntmax;
      plot(eee);drawnow;
  end 
  toc
%testing routine 
ntmaxTest=ntmax;
for nt = 1:ntmaxTest
    x = tpat(nt,2:ni+1);
    t = tpat(nt,1);
        neth = (wh*x')'+bh;
        h=1./(1+exp(-neth));
        neth1 = (wl*h')'+bl;
        k=1./(1+exp(-neth1));
        nety = (wo*k')'+bo;
    yy(nt,:) = 1./(1+exp(-nety));    
end
tpatt=[tpat,yy]
%plot xor input data to see  
% figure,plot(tpatt(2:3,1),tpatt(2:3,2),'*r')
% hold on,,plot(tpatt(1:3:4,1),tpatt(1:3:4,2),'o')
plot(yy,'o'); hold on; plot(tpat(:,1),'r*') 
%%%%%%%%%%%%%%%%%..................

3 个评论
显示 1更早的评论隐藏 1更早的评论

uday KUMAR 2012-2-24

Basically i supposed to use this code format..so if take some patience see my code...help mee through this hurdle i would be so glad to you....

Walter Roberson 2012-2-24

Using that code format does not prevent you from adding useful comments.

请先登录，再进行评论。

Help meeeee.....through this prblm.

2 个评论
显示无隐藏无

回答（2 个）

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

3 个评论
显示 1更早的评论隐藏 1更早的评论

类别

标签

Community Treasure Hunt

Help meeeee.....through this prblm.

2 个评论 显示 无 隐藏 无

回答（2 个）

1 个评论 显示 -1更早的评论 隐藏 -1更早的评论

3 个评论 显示 1更早的评论 隐藏 1更早的评论

类别

标签

另请参阅

Community Treasure Hunt

2 个评论
显示无隐藏无

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

3 个评论
显示 1更早的评论隐藏 1更早的评论