Quote:
Originally Posted by avinashsringeri
Hi Guys,
I have created a neural network trading system based on the strategies I used in manual trading. The profit and equity curves are amazing in the training.But on live data I am not able match the exact performance that I got with training set, but its not bad.
I have some questions.
1) How to avoid overfitting?
2) What is the right way of evaluating a neural net? While optimizing I get lot of good results I dont know which one to pick.
|
use the cross validation. In every epoch calculate the error of patterns that not exist in trainig set. When the error begin increase just terminate training process. It is simple.