Change of function used in RBF

Hi,

I am interested to know, how can I change the default Gaussian function of Radial Basis Function Neural Network to other functions such as "Cauchy, Multiquadric or inverse multiquadric.

It is mentioned in MATLAB help that it is possible to change the function via Model Setting, but I do not know where is the model setting. I am using a RBF code also nntool.

I will be appreciated if you guide me in this case.

Thanks in advance

Answer 1

[Greg Heath]

You misunderstood my answer.

Instead of using NEWRB which has limited flexibility, consider FITNET for regression and curvefitting or PATTERNNET for classification and pattern-recognition (Both call FEEDFORWARDNET). The catch is REPLACING

 net.layers(1).transferFCN = 'tansig';

with

 net.layers(1).transferFCN  = 'radbas';

to create a RBF with much more flexibility.

After establishing a subset of baseline performances with radbas, you can replace radbas with Cauchy or other functions for comparison.

This will be much, much easier than using NEWRB.

Hope this helps.

Thank you for formally accepting my answer

Greg

Answer 2

[Greg Heath]

I don't know exactly. However my first step would be to look at

 type newrb

and consider the consequences of altering

net.layers{1}.transferFcn = 'radbas';

and corresponding statements.

Next, I would look at

 type feedforwardnet

and consider the consequences of altering

net.layers{i}.transferFcn = 'tansig';

and corresponding statements.

Hope this helps.

Thank you for formally accepting my answer

Greg

[electronx engr]

Sir fitnet uses TRAIN command for regression. If we simply replace 'tansig' with 'radbas', doesn't it mean we are training radial basis network with backpropagation?

[Greg Heath]

1. Yes, it means that backpropagation is used.

2. Yes, the hidden layer functions are radbas

3. HOWEVER, each hidden layer high dimensional radbas function has a different shape and alignment in each projected input plane.

Hope this helps.

Greg

[electronx engr]

ok sir.. Thankx a lot