Module tutorial2

source code

tutorial2

Contains all parameters for the evolutionary run, grammar rules, constraints, and specifics about the terminal and function set of the trees in tutorial2. This example file gather all the settings for a simple polynomial regression using this time 2 variables x and y, and adding random integer constants between 1 and 5, and the following mathematical operators: '+','-','neg','*','^2'(or square). We try to find the following polynomial: x^2+3y+4 from 5 sets of testing data (5 different values for x and y). As Koza said, random constants are the "skeleton in Genetic Programming closet". Using them will definitely slow down a search... Any suggestions of current alternative strategies solving this problem would be highly welcomed :) Considering the constraints for building the trees, the root node will only have one child, and there will be no need for ADF in the function and terminal set. A typical way to run the tutorial would be to:

• modify the settings file so that:

  >>>     #setup for running tutorial 2
          functions = tutorial2.functions
          crossover_mapping=tutorial2.crossover_mapping
          nb_eval=tutorial2.nb_eval
          ideal_results=tutorial2.GetIdealResultsData()
          terminals =tutorial2.terminals
          Strongly_Typed_Crossover_degree=tutorial2.Strongly_Typed_Crossover_degree
          Substitute_Mutation=tutorial2.Substitute_Mutation
          treeRules = tutorial2.treeRules
          adfOrdered = tutorial2.adfOrdered
          FitnessFunction = tutorial2.FitnessFunction

• run the evolution in the main method, and make sure that the root node parameter only have one child.e.g.

  >>>    import evolver
  if __name__ == "__main__":
  >>>    dbname=r'C:\pop_db'
  >>>    evolver.EvolutionRun(2000,(0,1,'root'),2,8,'AddHalfNode',100, 0.00001 ,0.5,0.49,7,0.8,dbname,True)

  This means that we define:

• a population size of 2000 individuals,
• a root node with 1 child
• a minimum tree depth of 2
• a maximum tree depth of 8
• the use of a Ramped half and Half Koza tree building method
• a maximum number of runs of 100 generations before we stop
• a stoping fitness criteria of 0.1 (if the fitness<=0.1, solution found)
• a crossover probability of 0.5
• a mutation probability of 0.49
• a reproduction probability of 0.01 (automatically deduced from the 2 previous values)
• the size of the tournament selection
• the probability of selecting the fitttest during tournament selection
• a database of name and path dbname
• a run done in verbose mode (printing the fittest element of each generation)

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.