Multiple input feedforward network

I want to create a feedforward neural network with two input vectors and only one output vector. I've tried different things but I don't succeed.

I've tried:

P = [P1 P2];
net = newff(P,T);

But when I look at the architecture of the network it says:

numInputs: 1

Also when I try to train the network:

net = train(net,P,T);

I get the following error message:

??? Error using ==> network.train at 145
Targets are incorrectly sized for network.
Matrix must have 2 columns.

Error in ==> Problem1 at 90
net = train(net,P,T);

Is there anyone who can help me? I cannot seem to find a solution.

Maybe I should add: Both my input vectors (P1, P2) and also my target vector (T) have dimension [1000x1].

Jana

Answer 1

[Greg Heath]

>I want to create a feedforward neural network with two input vectors and only one output vector. I've tried different things but I don't succeed.

>I've tried:

>P = [P1 P2];net = newff(P,T);

If P1 and P2 are 1-D column vectors, then you need a lot more data.

For an I-H-O feedforward net, with

[I N] =size(P)

[O N] = size(T)

Neq = N*O % No. of training equations

Nw = (I+1)*H+(H+1)*O % No. of unknown weights

To mitigate measurement error, noise and differences between training and nontraining data, is desirable to have Neq >> Nw which is equivalent to

N >> Nw/H

or

H << (Neq-O)/(I+O+1)

>But when I look at the architecture of the network it says: numInputs: 1

That is correct. 1 input of dimensionality I=size(P,1)

>Also when I try to train the network:

>net = train(net,P,T);

This assumes a default of H = 10

>I get the following error message:

>??? Error using ==> network.train at 145 >Targets are incorrectly sized for network. Matrix must have 2 columns. Error in ==> Problem1 at 90 net = train(net,P,T); >Is there anyone who can help me? I cannot seem to find a solution.

>Maybe I should add: Both my input vectors (P1, P2) and also my >target vector (T) have dimension [1000x1].

See above for correct sizes for P and T and constraint on size of H.

Hope this helps.

Greg

[Jana]

I have trouble understanding your way of notating things.

What do you mean with [I N] and [O N]

I'm still altering the number of hidden neurons (and layers) in my network to find proper architecture.

Also I have transposed my vectors so now I have:

Input P [2x1000]

Target T [1x1000]

I hope this is the way to have two inputs in a network with one output?

[Greg Heath]

You have answered your own question for a feedforward net with I-H-O (Input-Hidden-Output) node topology.

size(P)

size(T)

Greg