Neural Network- IW doesn't fit with inputs

i have some problems with neural network toolbox.

1) i'd like to build a network with 9 inputs. when i generate the network (with 1 hidden layer and 1 output layer) by using neural network toolbox, i get an input vector of 1x8 matrix. It has to be 1x9, isn't it?

2) i need to generate an equation using my input parameters. that's why i use just one hidden layer. after generating the network i have input, layer weights and biases . when i use the equation given in the manual:

a3=f3(LW32*f2(LW21*f1(IW11*p+b1)+b2)+b3)

the output doesn't fit with the simulation result of same input. why does this happen and how can i reach the same solution with the simulation.

Answer 1

[Greg Heath]

1. Single inputs and outputs are column vectors. Orient your input and target matrices so that they have the same number of columns.

2. A single hidden layer is sufficient. I do not recommend using more unless you have a concrete specified reason for doing so.

3. Unfortunately, the creation/configure/train functions contain transparent default operations. The ones that are probably affecting your comparison are:

    a. Normalization and Unnormalization
    b. Random data division into train/validation/test subsets.

3. Either disable the defaults or take them into consideration.

Hope this helps.

Greg

[Turev]

sorry i have mistaken while writing the problem.

my input matrix is 9x183 and my target matrix is 3x183.

the problem is i have an IW of 1x8 vector.

i'm new in this subject. that's why i'm using the toolbox instead of writing code.

i'll try to use your recommendations (if i can)

Answer 2

[Greg Heath]

Searching the Newsgroup using IW LW

http://www.mathworks.com/matlabcentral/newsreader/view_thread/318575#872106

Hope this helps.

Greg