Genetic programming problem

[Maxim]

Hi everyone.

I have downloaded HeuristicLab 3.3 and tried to use it to model a very
simple genetic programming problem. Unfortunately I could find any
documentation or tutorial explaining how to do this.
I need to create an optimizer which would use custom defined functions
to find a compound function to fit the existing data set. The function
set will contain functions which can operate on numbers and lists of
numbers.
I did check the user manual and tutorials, but I'm still puzzled as
where to start.

Thank you,

[Gabriel Kronberger]

Dear Maxim,

in HL it is not possible to define new functions with specialized behavior
for GP in the GUI. This is a use case where it is necessary to write source
code implementing additional classes to get the desired effect. You can look
at the source code of the artificial ant problem
(http://dev.heuristiclab.com/trac/hl/core/browser/trunk/sources/HeuristicLab
.Problems.ArtificialAnt/3.4) to get an idea how to implement the
functionality you need.

At least the following things must be implemented:
* Symbols: each function and terminal symbol that you need for your problem
must be represented as a separate symbol class.
* Grammar: the grammar defines which functions are allowed as arguments of
other functions. This allows to restrict the syntactical structure of your
programs for instance functions that take lists of numbers can only be
applied to functions or terminals that produce a list of numbers.
* Fitness evaluator: takes a symbolic expression tree made up from the
custom function and terminal symbols and produces a quality for the given
tree.
* Problem class: holds additional parameters of your problem and the
operators.

I can help you further, if you post more detailed information about the
problem you want to solve.

Kind regards,
Gabriel