Is there any RBF neural network function available for classification problem?

Hi,

As far as I know, the neural network pattern recognition tool does not supply the radial basis network algorithm. Is there any RBF neural network function available in Matlab for classification problem?

Many thanks!

Betty

Answer 1

[Greg Heath]

help newrb

doc newrb

Search the newsgroup using

heath newrb

Hope this helps.

Greg

[Moore betty]

Thanks for your help.

As for function newrb, it seems that it is used only for approximating a function in Matlab , do you know if the newrb function could be used to classify a dataset? Since I am new to neural network, hope you could give me some suggestions. Many thanks!

Betty

Answer 2

[Richard]

Hi - I think the free Netlab toolbox has code for radial basis function networks: http://www1.aston.ac.uk/eas/research/groups/ncrg/resources/netlab/

[Moore betty]

Thanks for your information!

I have download the toolbox, it has many useful codes though I still cannot find how to use rbf network for classification.

Thank you all the same!!

Best Regards

Betty

Answer 3

[Greg Heath]

108 results for

heath newrb classification

Hope this elps.

Greg

[Moore betty]

Hi, thanks for your help!

I have searched for your answer, I still have a question that PCTerrval (the validation data missclassification error rate) you mentioned as performance goal for classification, cannot be found in Matlab. Does that mean I have coding this part in function newrb by myself?

Hope for your reply!

Many thanks!

Answer 4

[Greg Heath]

Yes.

1. Divide the data

2. Normalize or standardize the training set

3. Use the training set statistics (min/max or mean/std) to normalize the trn and val sets.

4. NEWRB does not consider val or tst data. Training stops when either Hmax or MSEtrngoal is reached. I typically use Hmax <= (Ntrn-1)*O/(I+O+1) and MSEtrngoal ~ 0.01*mean(var(t'))

5. Train multiple designs with different values of spread

6. Record the trn/val/tst MSEs and PCTerrs.

7. Choose the design with the best nontraining performance

8. Searching with heath newrb close clear should yield code examples.

Hope this helps.

Greg