multiple input/output problem in ANN

So I have 1000 cases, in each case i have the input parameters with 10 samples and 9 element (10X9)like a cell array with this form:

P =

      {1x9  cell}
      {1x9  cell}
      {1x9  cell}
      {1x9  cell}
      {1x9  cell}
      {1x9  cell}
      {1x9  cell}
      {1x9  cell}
      {1x9  cell}
      {1x9  cell}

and {1X3} =output cell array for each case, I am trying apply the inputs/outputs in nftool, but its showing me sample mismatch error. Can you please suggest how to work with multiple inputs/outputs in ANN?

Answer 1

[Greg Heath]

You can use NEWFF or NEWRB. The former is probably easier.

I do not understand the last sentence because you are not using proper terminology.

1. Reread my comments on standard terminology

2. Rephrase your question in terms of the input and output size formulas that I gave.

[Greg Heath]

N = sample size

I = input vector dimensionality

O = target and output vector dimensionality

Hope this helps.

Greg

Answer 2

[Greg Heath]

Your notation is confusing. Conventional statistics terminology considers a sample as a subset of a population of multidimensional measurements or cases.

To design a FFMLP with I-H-O topology, the data sample contains N pairs of I-dimensional inputs and O-dimensional target outputs stored in input matrix, p, and target matrix, t:

[ I N ] = size(p)

[ O N ] = size(t)

I am guessing that, for your problem

N = 1000

I = 90

O = 3

Is that correct? If so, just see the examples in the documentation. However, I suspect that you can characterize your inputs in vectors with much smaller dimensions.

For regression, consider PCA (Principal Component Analysis) for dimensionality reduction whereas PLS (Partial Least Squares) is probably better for classification.

Hope this helps.

Greg

Answer 3

[Tomaloka Chowdhury]

Hi Greg,

Thanks for your reply. Can you please provide me link for the documentations on FFMLP with I-H-O topology? I am a beginner in ANN, I dont have much concept of this. I want to make sure with you that input/output sample size dont need to match in ANN, as I am having 10 sample input with 9 element that will fit to 1 sample of input with 3 element.If you can provide me similar kind of code, that will be very helpful.

Also,I am using matlab 7.6.0.324 version.Is it sufficient to simulate this program?Please let me know. Thank you so much for your co-operations.

Regards, Tchow