Feature Request - pow (x,n)
  • Hi.

    Please define this function: pow(x,n) which n is an float number.

    In almost all cased, the pow(x,y) uses another variable for the power, not a float number. 

    Thanks in advance.

  • Thank you for your suggestion. Implementing this function would require a different, more constrained learning algorithm and changing the GEP-RNC algorithm to accommodate this is neither feasible nor sensible (evolution would become much more inefficient). As it is, the GEP-RNC algorithm fine-tunes the inputs to the power function totally unconstrained, choosing among all possible combinations. Notwithstanding, you can prime evolution with a seed model in which you fix the exponent to be a constant. Then you can create a Modeling Strategy (by choosing certain Genetic Operators) where you restrain evolution around your fixed structure.

    Candida Ferreira

  • Dear Dr. Ferreira,

    Thanks for the reply and suggestion, but fixed exponent is not desirable as the aim is finding the best exponent which fits data.

    My suggestion was implementing new power function, not limiting the current one. We could have two functions:


    IntPow (x,n)

    The second one is more useful in more cases and exists in other software applications. 

    For example in viscosity as a temperature dependent property, it can be defined as a f ( DP, pow(T,n)) (DP stands for dipole moment); not pow(T,DP), which the latter is done by GeneXproTools.

  • Hi Mehdi,

    The exponent is not fixed: what is fixed is the structure; the value of the exponent will be fine-tuned by the learning algorithm in order to fit the data.

    You can implement easily an IntPow(x,n) function in GeneXproTools as a custom function (DDF). Notwithstanding, when you use the built-in Pow(x,y) the learning algorithm will discover an optimal configuration to fit the data and you, as the designer, can have the final say and select the models that are making use of functions with the form you find most appropriate. Then you can use these models as seed to create even better models with these characteristics.

    Hope this helps.

    Candida Ferreira

Howdy, Stranger!

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