How to incorperate graph as nerual network input?

Hi, I'm new to using neural network in Matlab. At the moment I wish to use the graph results collected from my experiment as inputs to the network for pattern recognition. Question is can the neural network read the graph data as inputs? and how should I put it in the network. Thanks

Answer 1

[Greg Heath]

The net reads matrix input only. For a MLP with I-H-O node topology the input and target formats are

[ I N ] = size(input)

[ O N ] = size(target)

Hope this helps.

Thank you for formally accepting my answer.

Greg

[Jesmond]

Thanks a lot Greg, this has increase my understanding.

This lead to another question, does the matrix has to be a square matrix? ie. 10 by 10?

Well, my experiment results are acquire from the oscilloscope. The file format is in .csv and the data import into Matlab is in a 4000x1 matrix.

[aurora]

I have the same question , and I want to know what's (I) and what's (N)in the size of input and taregt ? Do the columns N express the no. of classes or the different inputs of the same class or the length of input elements ?

A numerical example will clear the problem .. if I have 10 different classes , each class has 10 shapes to train the network on them , a feature vector of each shape is extracted and the all ten feature vectors of the same class must be trained and produce the same target, the same process is done for all the classes. How to format the input as a suitable matlab neural network toolbox input?

[Greg Heath]

Input vectors and target vectors occur in pairs but have different dimensions. When all of the data is collected into separate input and target matrices, the dimensions of the matrices are as I have given previously. Each I-dimensional input vector has an O-dimensional output vector as a training target.

For classifiers O = c, the number of classes, and each target vector is a column of the c-dimensional unit matrix eye(c). The index of the row containing the one is the class index of the corresponding input vector.

Jesmond: It is not enough to give dimensions of a data matrix. You have to give dimensions of paired input and output vectors.