How to give input for back propagation program using ANN? Can we give a matrix of some values? And how do we specify hidden layers like on which basis we have to choose number of hidden layers?

1. I want to classify some values for that i want to use ANN 2. According to my study for classification back propagation and probabilistic NN are some good methods. 3. So want some help to choose the best method among them. 4. I have a matrix of some features such as mean, variance..using that want to do classification. 5. Now how to do classification using different ANN & using Neural network toolbox in MATLAB

[vaishnavi kaundanya]

Ok thank you so much. Shall I get the same help for probabilistic neural network in a neural network toolbox?

[Greg Heath]

 Sure. See the help and doc documentation and search for PNN and NEWPNN posts.

Answer 1

[Greg Heath]

patternnet is a universal approximator.

1. Start with the examples in

 help patternnet
 doc patternnet

2. See some of my posted designs in the NEWSGROUP and ANSWERS

 greg patternnet

3. Continue with other MATLAB example data

 help nndatasets
 doc nndatasets

Recommendations

 1. Standardize inputs to zero-mean/unit-variance
2. Plot inputs for each class
3. Remove or modify outliers
4. Initialize the RNG to your favorite state
5. Use default settings and design 10 nets (for i = 1:Ntrials) with different random data divisions and initial weights
6. If successful, try making the net more robust by reducing the number of hidden nodes.
7. Otherwise (if unsuccessful),  try increasing the number of hidden nodes.

Hope this helps.

Thank you for formally accepting my answer

Greg

[Greg Heath]

Rather than use NEWPNN, use NEWRB

 help newrb
 doc newrb

Search

 greg newrb

Hope this helps.

Greg