How to pre-process training data, the x-y coordinates data represent the Gaussian graphs for ANN training?

Hi all, I am trying to train a network to produce the Sigma value and Delta value of the Gaussian & Dual-dirac delta. The inputs is the x-y coordinates of the Gaussian plots. every plot represent certain value of Sigma & delta.

Now due to the range of Sigma & delta, there're a lot of redundant x-y pairs and causing a lot of noise for the training data.

I am trying to just extract the enough "meaningful" N numbers of x-y coordinates per plot to feed to the ANN for training. and still need to maintain the same training data array dimensions for every plots.

For eg: To generate M numbers of plots for the training (i have this function ready), and N x-y pairs per plots The input training data array dimension will be NxM. the Target training data array dimension will be 2xM.

Example of the plots as follows: and the Input training Raw data of the X-Y coordinate are stacked as the attached spreadsheet which consists of many "near to Zeros" coordinates. </matlabcentral/answers/uploaded_files/126265/gauss.PNG>

My current problem are:

1. How to pre-process the training data or sample the meaningful plots with same N number of X-Y coordinates, for a M numbers of plots (may have few hundreds plots)? PS: I have the function to create the plots, the X-Y data is pre-generated to create the plots. I hope I have stated the question clearly. if not, please do ask me again.

Thank you very much.

Answer 1

[Bernhard Suhm]

Have you tried using multcompare on the stats object that aoctool delivers, like alluded to in the doc page for aoctool ?

[Chin Kui Ku]

Oh, this is something new to me, will try it out. thanks.