Hello every, I'm a student and I am desperately in need of help in order to complete my final year project (less than a month remaining).

I have to design a backpropagation NN using matlab (6.5). My input consist of 100 components. I know this a large figure and i've tried using pca to reduce the dimensionality. i managed to obtain a a 7 component transformation matrix. However I had to do same hacking b'cos matlab would complain about some issues with the matrix dimensions.So I think this result is wrong.

Anyway out of the way, my main purpose of this post is that i need help to create a 100 element backpropagation NN. I bhave generated a training set of 7 100 elements. I am using the neural net editor (nntool) to create the network. I created a 2-layer net and also import a 100x1 input to apply to the net. As this represents a single input(with 100 components) i create a target by making a 1-element target matrix. When it comes time to train the net, I always encounter this message:
"Error using ==> network/train.
Inputs are incorrectly sized for network.
Matrices must all have 2 rows."

I have tried different configurations but to no avail.

Prior to using nntool I read the documentation for the Neural network toolbox to prepare myself. I am feeling very lost and have just under a month to complete my project. Please, Please,Please help.

P.S.: sorry this post was so long.

Hello,I think I can give you some help.

The problem is in the sizes of the inputs and targets. This have to be (input_length*inputs_number) and (targets_length*inputs_number).

Then,when you create the new network,change the default input range [0 1;-1 1] selecting Input Ranges: Get from input=Your_input_matrix_name.

Good work.

Hi There,

I have the same problem, but I'am afraid I do not uderstand this explanation. Could you please (or somebody else) be more specific?

What if I have a 5x1 input vector, do I have to create a 5x5 input matrix and fill it with zeros or alike?

Many thanks.