Evolutionary Parameter Optimization setting

Legacy User

RapidMiner has the nice feature to optimize parameters using a
genetic algorithm. I would like to play around with this optimization.
However, I'm not sure which parameters I should use for the operator
EvolutionaryParameterOptimization (no, I don't want to train
automatically by RM:-)). There are plenty of different settings and
I'm unsure which combination I should use to achieve best results.

Do you have any general hints for me that I should take into account?

Moreover, do you know any papers/books/websites that discuss the
topic of selecting machine learning parameters via genetic
algorithms? The only paper I've found so far is:
¨Comparing learning approaches to coreference resolution".

Regards,
ben

land

Hi Ben,
if you address the parameters for selecting the evolutionary strategy:
The main parameters are number of individuals and the maximal number of generations. The other parameters influence the number of generations needed to converge (and hence the computing time), but in any setting it should converge anyway. Which settings work best depends on the problem.


Greetings,
  Sebastian

Daniel_Zukowski

Hello Ben,

I suggest you the following Google Scholar searches:
http://scholar.google.com/scholar?hl=en&;lr=&q=data+mining+genetic+algorithms
http://scholar.google.com/scholar?hl=en&;lr=&q=machine+learning+genetic+algorithms
For example, see this paper:
A survey of evolutionary algorithms for data mining and knowledge discovery

Also, there is a good free book about optimization: Global Optimization Algorithms.

Regards,
Daniel

steffen

Thanks for the links, Daniel !