Neural Networks, Custom performance function

Hello,

How do you implement a neural network with a user defined performance function?

Say I have created a function AsymPerformance.m :

------

funtion ret = AsymPerformance(y,yHatFromNN)

ret = (y > yHatFromNN) * abs(yHatFromNN - y)' .* 2 + (y < yhatFromNN) * abs(yHatFromNN - y);

end

-------

How do I train my NN to optimise minimising this as a measure of performance?

Please try to keep it simple, new to NN..

Thank you in advance!

Kris

Answer 1

[Kristoffer Uppheim]

Found the answer:

http://www.mathworks.com/support/solutions/en/data/1-2VFDSE/index.html?product=NN&solution=1-2VFDSE

Sry for double posting..

[Omer]

It is little complicated do understand. There is no easy way?