Username Remember Me?
Password   forgot password?
   
   
multi classes many objects at one class each in dataset ,how to label and train with backprop neural network
Posted: 16 August 2009 10:43 AM   [ Ignore ]  
Novice
Rank
Total Posts:  18
Joined  2009-08-13

my data set consest of 25500 letter image to 34 diffrent
letter each letter has 750 image stored in
one folder as training dataset k and 6800 letter image 200
for any letter as testing dataset
25500 training images 30x30pixel ,350 features

the input = 350 feature

classes = 34 ( the number of letters)

the classifeir is backpropagation neural network
load

d= data.X;

t=data.y’;

net=newff(minmax(data.X),[350,34],{’tansig’,’purelin’});
net.trainParam.show =600;
net.trainParam.lr = 0.05;
net.trainParam.epochs = 300;
%net = init(net);
%training
[net,tr]=train(net,data.X,data.y);
my problem

i want the 34 output node ( the number of classes )
but the matlab error message is ( the matrix must have 34 rows)

dataset , data.X= (350 25500)

labels target , data.y = ( 25500 1)

i want code with backpropagation using matlab
to classify this data set and test iT.

[ Edited: 15 September 2009 09:07 PM by omerneural]
Profile
 
 
Posted: 17 August 2009 09:46 AM   [ Ignore ]   [ # 1 ]  
Novice
Rank
Total Posts:  18
Joined  2009-08-13

no one know my problem i want my code doing with 34 output nodes
34 class to 25500 images in one matrice with matlab software

Profile
 
 
Posted: 18 August 2009 06:45 AM   [ Ignore ]   [ # 2 ]  
Novice
Rank
Total Posts:  18
Joined  2009-08-13

hoooo my friend

are you dont want me with your forum

why no one replyyyyyyyyyyyyyy?????????????????????

Profile
 
 
Posted: 18 August 2009 08:58 AM   [ Ignore ]   [ # 3 ]  
Novice
Rank
Total Posts:  15
Joined  2009-07-15

Hi

> net=newff(minmax(data.X),[350,34],{’tansig’,’purelin’});

This actually inits a 3-layer network, with TWO hidden layers (350 and 34). You also do not need 350 neurons in the first hidden layer. Try something like this:

net=newff(minmax(data.X),[8],{’tansig’});

Profile
 
 
Posted: 18 August 2009 06:24 PM   [ Ignore ]   [ # 4 ]  
Novice
Rank
Total Posts:  18
Joined  2009-08-13

thank marty
but net=newff(minmax(data.X),[8],{’tansig’});
8 is output node
my problem still

[ Edited: 18 August 2009 06:28 PM by omerneural]
Profile
 
 
Posted: 18 August 2009 07:14 PM   [ Ignore ]   [ # 5 ]  
Novice
Rank
Total Posts:  15
Joined  2009-07-15

my fault, it should be:

net=newff(minmax(data.X),minmax(data.y’),[8],{’tansig’});

Profile
 
 
Posted: 20 August 2009 01:22 PM   [ Ignore ]   [ # 6 ]  
Novice
Rank
Total Posts:  18
Joined  2009-08-13

my teacher
no solution
any matrix of image in dataset has a label
many matrices of images to one class
output node in last layer must be equal to
the classes number
the label to my dataset is data.y=(25500 1)
25500 matrix represrnt 25500 image
to 34 class any class has 750 image matrix

Profile
 
 
Posted: 20 August 2009 02:26 PM   [ Ignore ]   [ # 7 ]  
Novice
Rank
Total Posts:  15
Joined  2009-07-15

You have 25500 objects with 350 features each and you want to map it into 34 classes, right?

You need a network with 350 input units, some number of hidden units and 34 output units. To create such network you have to issue the following command:

net=newff(minmax(data.X),minmax(data.y’),[8],{’tansig’});

The number of input and output units will be determined automatically from minmax(data.X) and minmax(data.y’). You only need to provide the number of hidden units, 8 in this case.

Now, your problem is that data.y = (25500 1), it should be (25500 34) - you have to convert it from class label (1..34) into a binary target vector, otherwise it will not work.

Check MATLAB documentation, it’s all there.

Profile
 
 
Posted: 20 August 2009 03:50 PM   [ Ignore ]   [ # 8 ]  
Novice
Rank
Total Posts:  18
Joined  2009-08-13

helo my friend Marcinbu
oh how i can do label (25500 34)into binary target vector?
my code to convert images pattern to vector :
any class images=750 sample store in one folder
and i have 34 folder
code like this
for allFolders=1:34

SubIm=reshape(SubIm,1,[]);

?d reshape image to image matrix

X_vector = [ X_vector ;SubIm ];

?d reshape image’s class to classes matrix

y_vector = [ y_vector; ones(1,1)*(allFolders-1)];
how i can do classes 34 in this code

[ Edited: 20 August 2009 04:06 PM by omerneural]
Profile
 
 
Posted: 26 August 2009 11:13 PM   [ Ignore ]   [ # 9 ]  
Novice
Rank
Total Posts:  18
Joined  2009-08-13

my friends
how wwwwwwwwwwwwwwwwwwwwwwwwww
i can convert large dataset image contains
34 class each class has 700 image
to matrix
SubIm=reshape(SubIm,1,[]);

?d reshape image to image matrix

X_vector = [ X_vector ;SubIm ];

?d reshape image’s class to classes matrix

y_vector = [ y_vector; ones(1,1)*(allFolders-1)];

Profile
 
 
Posted: 27 August 2009 12:01 AM   [ Ignore ]   [ # 10 ]  
Novice
Rank
Total Posts:  18
Joined  2009-08-13

hep meeeeeeeeeeeee

Profile