Posted By: hersh   Date: Wed Jul 9, 2008

Title: Curve fitting is OKAY...

Message:
  Thanks for your continuing effprts. I use these with e-wave and other indicators and it adds "confidence" to contrarian trading.
  Re: criticisms about back-fitting... we all project future direction from past action. There is no other method. I believe it is reasonable to assume that the underlying structure of the market/economy changes over time too. Hence I would even recommend ROUTINELY reruning the "AI" networks perhaps every 3 months to adapt to the "unseen" underlying structural changes. (Note- I'm not a systems engineer or mathematician or game theorists etc etc so I'm speaking as a layman whose observed markets and trading systems for several decades)

Response:
  I hardly ever retrain or rerun existing networks. I do replace networks in the library when I see that they are not performing as well as I would like. Retraining only adjusts the parameters (weights) of the networks. It leaves their equation structure unchanged. Generating new networks allows the equations to be reformulated to reflect current market conditions.

  Version 16.2 contains a new set of 222 networks. We will add nets when the genetic algorithm can find better individuals. I have 7 CPUs running full time to find better nets but the dimensionality of problem is so large, it could take weeks before one is found that passes our validation tests.