Best Fitness value
  • Is there a specific goal I should be modeling towards for the "Best Fitness" value  when using Max Fitness in my modeling? I'm a little unclear

    how I can use that value and what I should be targeting / looking for - I'd like to be able to say, Oh, that's a good number or..hum...I think I need to hit the Continue Button again...

    thank you guys!


    Devon Kyle

  • That’s a hard one, Devon. It depends. First of all, it depends on what you need and how fast you need it. It also depends on the modeling category.

    If you’re creating time series prediction models and you’re dealing with financial time series such as stocks, I don’t think you’d benefit much by hitting Continue again and again. I’d rather analyze the intermediate models or create several independent runs, keeping and appending just the best of each run to the history, then select the best one from these.

    If it’s regression or classification/logistic regression, I usually try to do better than simple standard techniques such as the multilinear regression model or the canonical logistic regression model, which is not hard with GeneXproTools. For example, the multilinear regression model for the Concrete Strength data used in the Concrete Strength sample run has an R-square of about 0.6; with the same data in GeneXproTools you can create good (in the sense that they generalize well) nonlinear models with 0.87. You’ll have to hit the Continue Button a few times for that, but you can do it; it’s just a matter of how much time you have and how soon you need your models.

    And it also depends on the final application of the model: is it going to be a standalone model or part of a model ensemble? We’ve introduced new stop conditions in v5 to help with that, where you can specify a random value (for the fitness, R-square, accuracy, or generations) between two values in order to generate a diverse bunch of models with different, and hopefully complementary, properties.


  • As always ,that is a big help Candida

    Hope others can find this info useful as well.


    ... Gracias!


Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!