How can I pass the performance parameters to my own custom performance function in the Neural Network Toolbox?

How can I pass the performance parameters to my own custom performance function in the Neural Network Toolbox?

Answer 1

[MathWorks Support Team]

For releases of MATLAB prior to MATLAB 8.0 (R2012b), the MSEREG function within the Neural Network Toolbox is a good example of a performance function using the performance parameters.

1) The custom performance function should have a header that looks like:

function perf= myperfFcn(e, x, pp)

where:

 E  - Matrix or cell array of error vector(s).
 X  - Vector of all weight and bias values.
 PP - Performance parameter structure.

2) To specify your custom performance function:

net.performFcn = 'myperfFcn'

3) To specify your custom performance parameters:

      perfStruct.varA = 1;
      perfStruct.varB = 2;

      net.performParam = perfStruct;

Note the "net" structure will now include:

    parameters:

         adaptParam: .passes

         initParam: (none)

         performParam: .varA, .varB

For releases of MATLAB starting in MATLAB 8.0 (R2012b), information is provided by the command,

>> help nncustom

For example, to specify a custom performance function, execute the command,

>> which mse

to locate the 'mse' function.  Copy and rename the file 'mse.m' and the folder '+mse'.  Then, use these functions as templates to write your own custom function.