Neural Networks manipulation in k fold method

so after using the k-fold method (for validating and testing each subset K times) is there a way to manipulate the k "subnetworks" created? i there a way to make these k networks visible and accesible? Is my question meaningfull? i mean what happenes in k-fold is creating k networks or not?

Answer 1

[Greg Heath]

The simplest solution is

y = mean( net1(x)+net2(x)+...netk(x));

Any effort to combine weights into one net has to take into consideration the different default normalizations. Therefore, all of the data would have to be standardized or normalized a priori using the same mean/stdv or min/max and the default normalization disabled.

Hope this helps

Thank you for formally accepting my answer

Greg

Answer 2

[laplace laplace]

let me re-phrase my question to make it more clear. Can i use each of the k-networks created independently from the others?

[Greg Heath]

Yes. Test on all of the data and choose the best.

[laplace laplace]

is there a command to do so?

[Greg Heath]

No special command. Just treat it as as if you only trained one net.

[laplace laplace]

what is the argument "x"

y = mean( net1(x)+net2(x)+...netk(x));  if true

[Greg Heath]

All of the input data

Greg