Prev: how to Plot SVM results (libsvm) in 2d space. help needed.
Next: Matlab Surf
From: Fred on 8 Mar 2010 11:54
Hello everybody,

I'm experiencing exactly the same problem as Vladimir. I have a 786 sample set of near-infrared spectra. Each sample is characterized by 3112 variables (the spectra). So my data is a 786x3112 matrix.

According to literature, the number of neurons created should be around 33% and 50% of the total number of samples in the dataset. I have tried transposing the matrix (3112x786) which according to the examples shown in the Neural network toolbox guide (on the “Iris” dataset, http://www.mathworks.com/access/helpdesk/help/toolbox/nnet/getting7.html#27151), seems to be the correct way to do but the error will always be there.

Has anyone found a solution to this problem in the last 6 month?

Thanks for your help!

Fred
 | 
Pages: 1
Prev: how to Plot SVM results (libsvm) in 2d space. help needed.
Next: Matlab Surf