Neural nw : Inputs and targets have different numbers of samples

Hi All

I have a code , I am just checking how it works , my input matrice is :

input = [0.0600000000000000  0.00100000000000000  45  0.0508000000000000  0.0127000000000000]

and the target is a 6 by 6 matrix

so using this code bellow , I get the mentioned error : Inputs and targets have different numbers of samples ,

Error in Neural (line 17) , [net,tr] = train(net,xn_tr,yn_tr);

here is the full code :

clc 
clear 
clear all 
load('input.txt')
%load input 
load ('taget.txt')
%normalizing data 
[xn_tr,xs_tr] = mapstd(input); 
[yn_tr,ys_tr] = mapstd(taget); 
%%network 
net=newff(xn_tr,yn_tr,[7 7],{'tansig'},'traingda');%7 hidden tanh layer gradian descent adaptive 
net.trainParam.epochs =70; 
net.trainParam.lr = 0.05; 
net.trainParam.lr_inc = 1.05; 
%training network 
net.trainFcn='traingda'; 
[net,tr] = train(net,xn_tr,yn_tr); 
%randomizing initial value f weight matrix 
net = init(net); 
net.trainParam.show = NaN; 
u_t=mapstd('apply',x,xs_tr); 
%simulating output 
y_hat=sim(net,u_t); 
%plotting performance 
plotperform(tr) 
mse=mse(y-y_hat)

[farzad]

Also , for my above problem

can I use this code ? how should I change it for this ?

no_of_points=5000; %no. training points of each Set 
center1=[0,0]'; center2=[0,4]'; 
sigma1=1; sigma2=4; 

%randn('seed',10); 
p1=randn(2,no_of_points)*sigma1; 
p1(1,:)=p1(1,:)+center1(1); p1(2,:)=p1(2,:)+center1(2); 
t1= ones(1,no_of_points); 
p2=randn(2,no_of_points)*sigma2; 
p2(1,:)=p2(1,:)+center2(1); p2(2,:)=p2(2,:)+center2(2); 
t2= -ones(1,no_of_points); 

p = [p1,p2]; %training set (input) 
t = [t1,t2]; %training set (output) 

lr= 0.0001; %learning rate 

%Initialize 2-2-1 network 
rand('seed',0); 
% randn('seed',0); 
net=newff(minmax(p),[10,1],{'tansig','purelin'},'trainlm'); 
%net=newff(minmax(p),[10,1],{'tansig','purelin'},'traingdx'); 
net.trainParam.lr=0.1; 
net.trainParam.epochs=20; 
net.trainParam.show =50; 

net = train(net,p,t); %Batch learning 
a = sim(net,p); %Net Simulation 

a_01=a; 
ii=find(a > 0); 
a_01(ii)=ones(size(ii)); 
ii=find(a < 0); 
a_01(ii)=-1*ones(size(ii)); 
ii=find((a_01-t)==0); 
correct_classif_percent=length(ii)/length(t); 
fprintf(1,'Correct Classification = %5.2f %%\n',correct_classif_percent*100); 

% --- Data Representation --- 
%data preparation 
x=[-5: 0.5 :5]; 
y=[-5: 0.5 :10]; 
[xx,yy]=meshgrid(x,y); 
p_glob=[xx(:), yy(:)]'; % (input) 
a_glob=sim(net,p_glob); %simulate 

[ny,nx]=size(xx); 
a_glob_zz=reshape(a_glob,ny,[]); 

%plot results 
figure(1);clf; 
ii=find(t==-1); 
plot(p(1,ii),p(2,ii),'o'); 
hold on 
ii=find(t==1); 
plot(p(1,ii),p(2,ii),'.m'); 

[c,h]=contour(xx,yy,a_glob_zz,[0 0]); 
set(h,'LineWidth',2) 
set(h,'EdgeColor',[0,0 0]); 
axis image; 
%return; 

%plot results 
figure(2);clf; 
surf(xx,yy,a_glob_zz); 
hold on 
%mesh(xx,yy,a_zz); 
h=plot3(c(1,2:end),c(2,2:end),zeros(1,length(c(2,2:end)))); 
set(h,'LineWidth',3) 
set(h,'Color','k') 
title('Neural network Surface');

[Greg Heath]

It looks like you are trying to make a mountain out of a molehill.

1. [ 7 7 ] in newff specifies 2 hidden layers. One is sufficient

2. Where in the world did you get the number 7 from?

3. Why use the obsolete newff instead of the current patternnet?

4. If you don't have patternnet, why use newff instead of newpr for pattern-recognition?

5. Why are you specifying parameter values that are already have defaults?

6. If this is for practice, why aren't you using the MATLAB practice datasets?

 help nndatasets
 doc  nndatasets?

7. There are hundreds of successful classification posts in the NEWSGROUP and ANSWERS. Why not use one of those?

[farzad]

Dear and respected Dr. Greg

I really appreciate your very kind attention , honestly I have a really short time , too short to learn this coding step by step ,the next topic will be optimization for me to learn , I wish I could do it fast , I am reading the MATLAB help,pdf on Neural NW, well in the nndatasets ,I get a bit lost, where to start ,and after all , seeing a lot of different ways that might not be similar to these code , I should come back to this code to see what is wrong in it

question 2 : I got this code from a friend who has worked on another topic , and I was going to use it

3-I do not know the difference and why newff is obsolete here , I wish you would tell me

4- I get this as an advice, thank you

5- which one do you intend ?

6- it's a problem I should solve asap

7- well each question has it's own problem , I could not follow all , specially with the error that I get , I yet could not find an answer for that myself

[Greg Heath]

Dear and respected Dr. Greg

% I really appreciate your very kind attention , honestly I have a really short time , too short to learn this coding step by step ,the next topic will be optimization for me to learn , I wish I could do it fast , I am reading the MATLAB help,pdf on Neural NW, well in the nndatasets ,I get a bit lost, where to start ,and after all , seeing a lot of different ways that might not be similar to these code , I should come back to this code to see what is wrong in it

I do not recommend using this code. Start out with the code in

 help newpr
 and
 doc newpr

Then search the NEWSGROUP and ANSWERS using the search word newpr

 NEWSGROUP
       newpr         17 hits
       greg newpr     8 hits
 ANSWERS       
        newpr        68 hits
        greg newpr   56 hits

% question 2 : I got this code from a friend who has worked on another topic , and I was going to use it

Again, not recommended

% 3-I do not know the difference and why newff is obsolete here , I wish you would tell me

NEWFF is obsolete because MATLAB replaced it 5 years ago. Previously,

 NEWFIT for curve-fitting and regression. It automatically calls NEWFF
 NEWPR for pattern-recognition and classification. It also automatically calls NEWFF

>> help newpr

newpr Create a pattern recognition network.
Obsoleted in R2010b NNET 7.0.  Last used in R2010a NNET 6.0.4.

Similarly for newfit and newff.

=========================

Currently, MATLAB offers

 FITNET for curve-fitting and regression. It automatically calls FEEDFORWARDNET
 PATTERNNET for pattern-recognition and classification. It also automatically calls FEEDFORWARDNET

You can obtain the full online documentation of any function using help and doc in the command line. The biggest problem is that the corresponding examples tend to pretty Mickey-Mouse. However, they provide the basics from which you can start. On the other hand the scripts from the nntool are too detailed and do not emphasize the most important commands.

Finally, you can search the NEWSGROUP and ANSWERS for the examples in my posts. If you have any problems with them you can always alert me via email that you have posted a follow-up comment or question. I do not give direct advice via email.

% 4- I get this as an advice, thank you % % 5- which one do you intend ? % % 6- it's a problem I should solve asap % % 7- well each question has it's own problem , I could not follow all , % specially with the error that I get , I yet could not find an answer for % that myself

Finally, I suggest using patternnet instead of the obsolete newpr. However, if you want to go with newpr, see my suggestions above.

Hope this helps.

Greg

Answer 1

[Greg Heath]

Here is a simplified example using the NEWFF example in the help and doc documentation. I omitted

1. Using an inner for loop over multiple random weight initializations and data divisions. To see those type examples search on greg Ntrials
2. Extracting the individual trn/val/tst performances via using the training record tr to obtain the corresponding indices.

 % >> help newpr    
 % load simpleclass_dataset
 %      net = newpr(simpleclassInputs,simpleclassTargets,20);
 %      net = train(net,simpleclassInputs,simpleclassTargets);
 %      simpleclassOutputs = net(simpleclassInputs);

 close all, clear all, clc, plt = 0
[ x, t ]  = simpleclass_dataset;
[ I   N ] = size(x)   % [ 2 1000 ]
[ O N ] = size(t)    % [ 4 1000 ]
trueclass = vec2ind(t);
class1 = find(trueclass==1);
class2 = find(trueclass==2);
class3 = find(trueclass==3);
class4 = find(trueclass==4);
N1 = length(class1)    % 243
N2 = length(class2)    % 247
N3 = length(class3)    % 233
N4 = length(class4)    % 277
x1 =  x(:,class1);
x2 =  x(:,class2);
x3 =  x(:,class3);
x4 =  x(:,class4);
plt = plt + 1
hold on
plot(x1(1,:),x1(2,:),'ko')
plot(x2(1,:),x2(2,:),'bo')
plot(x3(1,:),x3(2,:),'ro')
plot(x4(1,:),x4(2,:),'go')

 Hub = -1+ceil( (0.7*N*O-O)/(I+O+1))  % 399
Hmax = 40   % Hmax << Hub 
dH    = 4   % Design ~10 candidate nets
Hmin  = 2   % I know 0 and 1 are too small

 rng(0)   % Allows duplicating the rsults
j=0
for h=Hmin:dH:Hmax
    j = j+1
    net            = newpr(x,t,h);
    [ net tr y  ]  = train( net, x, t );
    assignedclass  =  vec2ind(y);
    err = assignedclass~=trueclass;
    Nerr           = sum(err);
    PctErr(j,1)    = 100*Nerr/N;
end
h = (Hmin:dH:Hmax)';    
PctErr = PctErr;
results = [ h PctErr ]

Hope this helps.

Thank you for formally accepting my answer

Greg

[farzad]

I really Appreciate your help

thank you so much

[farzad]

Hello Dear Proff. Heath

I wanted to try this code with the Sine(t) function , by defining t in a range [-2::0.1:2] , but the code gave me some matrix dimensionality problem

[Greg Heath]

The newPR code is for PatternRecognition and classification. Use newFIT for FITting the sine function.

[farzad]

Dear Professor Heath

one curiosity, why don't you write a paper or book, explaining your method here ?