Proper GRNN use in Matlab
in [Matlab]
|
From: Nathan on 1 Jun 2010 14:53 I have been looking at metal properties with respect to it's life, and trying to predict likely ranges that the life would fall into. I discovered neural networks, and it seemed that they would prove useful in my efforts. The most useful seems to be the GRNN(general regression neural network), with it's use in regression and generalized point, and I found the newgrnn function in Matlab's Neural Networks package. However, I was not able to find any more than the function itself, no guidelines on how to train or implement it. If anyone knows how the code might be structured, it would be greatly appreciated. The properties(inputs) are held in a 9x111 matrix and the lifespans(outputs) are in a 1x111 matrix. Anything would be helpful.
From: Greg Heath on 5 Jun 2010 14:47 On Jun 1, 2:53 pm, "Nathan " <nsjacks...(a)gmail.com> wrote: > I have been looking at metal properties with respect to it's life, and trying to predict likely ranges that the life would fall into. I discoveredneuralnetworks, and it seemed that they would prove useful in my efforts. > > The most useful seems to be the GRNN(general regressionneuralnetwork), with it's use in regression and generalized point, and I found the newgrnn function in Matlab'sNeuralNetworks package. However, I was not able to find any more than the function itself, no guidelines on how to train or implement it. > > If anyone knows how the code might be structured, it would be greatly appreciated. The properties(inputs) are held in a 9x111 matrix and the lifespans(outputs) are in a 1x111 matrix. Anything would be helpful. GRNN is constucted from all of the data. The only training is to determine the Gaussian spread by trial and error. My recommendation is to use either NEWFF and/or NEWRB doc newff help newff greg heath pre training advice doc newrb help newrb ..greg heath rbfnn training Hope this helps. Greg
|
Pages: 1 Prev: Event driven simulation Next: moving average code taking too long |