newff classifying

[EDIT: 20110624 15:42 CDT - merge duplicates, reformat - WDR]

Mr Greg

i saw so many answers given by you but still i getting problem with "newff" function

.............please help me anyone regard this.....................

1) i am constructing a network which is

   net1=newff(minmax(I'),[79 4],{'tansig' 'logsig'},'traingdm');   

    size(I)=200*20

   I is the set of 20 images

   size(T)=4*20

   i.e.,i want to classify the images into 4 classes

2) while i am training i getting the error like this

   net1=train(net1,I,T)
   ??? Error using ==> network/train
   Inputs are incorrectly sized for network.
   Matrix must have 20 rows.

what is the problem with this...............?????

please explain

Please help anyone

   regards
   *vijay*

[Material from duplicate #1]

dear friends i am facing problem with the newff function in the process of constructing the MLP...........

PLEASE HELP ME FRIENDS I AM DOING MY ACADEMIC PROJECT WORK ON CLASSIFICATION AND i AM FACING THE FOLLOWING PROBLEM

>> size(I)

   840    10

>> size(T)

       5    10

>> net6=newff(minmax(GPCOM),[79 5]);

>> net6.trainParam.epochs=5000;

>> net6.trainParam.goal=0.00001;

>> net6=train(net6,I,T)

??? Error using ==> network/train

Inputs are incorrectly sized for network.
Matrix must have 10 rows.

thanks friends

regards

VIJAY

[Material from duplicate #2]

dear friends i am facing problem with the newff function in the process of constructing the MLP

........... PLEASE HELP ME FRIENDS I AM DOING MY ACADEMIC PROJECT WORK ON CLASSIFICATION AND i AM FACING THE FOLLOWING PROBLEM,

Please help me so that i can complete classification......sincerely i need it .....

>> size(I)

   840    10

>> size(T)

       5    10

>> net6=newff(minmax(GPCOM),[79 5]);

>> net6.trainParam.epochs=5000;

>> net6.trainParam.goal=0.00001;

>> net6=train(net6,I,T)

??? Error using ==> network/train

Inputs are incorrectly sized for network. Matrix must have 10 rows.

thanks friends

regards

VIJAY

Answer 1

[Walter Roberson]

Pass T.' to train() instead of T.

(Internal note: newff does not exist in the newest NN toolbox and it is hard to find the doc page for it. The doc page exists at http://www.mathworks.com/help/releases/R2010a/toolbox/nnet/newff.html )

[vijayasarathi japa]

but sir

here i have 10 inputs and output should be classified into 4 classes

i saw in many posts that all inputs and corresponding outputs should be in columns

so if i transpose the target(T) then the target will differ which i need

[Walter Roberson]

Please file a bug report with MATLAB Technical Support.

[vijayasarathi japa]

thank you sir for your response

sir i successed in running the command"newff" by the previous matlab answers & i tried like this

net=newff(minmax(input),[79 5]);

net=train(net,input);

no of classes that to be classified are 5

here "input" is set of 25 images which containing 5 images for each class which are placed in columns(one single image represents 1 image)

But after training the network, i test the network using two images but i didnot get the correct answer....

please tell me how can i improve the network performance