On May 14, 2012, at 5:32 AM, Longdexin wrote:
> 1. I see so many constants in a formula when using the software Eureqa and
> they can change themselves to make the formula better and better. How can
> I do the same tricks with ECJ ?
This sounds like an algorithm or technique question. ECJ does straightforward implementation of genetic programming algorithms but I cannot say how to implement a specific custom algorithm like used in Eureqa. I can say that if you're trying to mutate constants, see the MutateAllERCs and MutateOneERC breeding pipelines.
> 2. I have known that GP may have overfitting problems. Is there anything can
> be done to avoid overfit as much as posisible in ECJ, like doing a k-fold cross
> validation ?
Again, this is an algorithmic question. ECJ does genetic programming -- perhaps you might ask the genetic programming user group (on yahoo groups) about how to get GP to do such things in general.