Questions related to Breeder

[Xiaodong Li]

Hi Marcus

I was testing the Breeder 2.1 with a multi-modal function De Jong's F5, 
then I noticed that the following code did not always give the minimum 
fitness value from the popList (as you can see from the plotted graph 
"Fitness Evolution", the minimum fitness has gone up close to the 
maximum fitness very quickly only after generation 40):

-------------------------------------------------------------
 [fitnessGraph createMinSequence: "Min fitness"
                withFeedFrom: [populationSwarm getList]
                andSelector: M(getFitness) ];
-------------------------------------------------------------

However if I add a line of code (shown below) to the "-evaluate" in 
popSwarm.m, to see each member's fitness:

//_______________________________________________________________
-evaluate {     // Evaluates all member of the population
//_______________________________________________________________

// Takes them from the new list and places then in the Population map,
// keyed with fitness

    id index, member;
    index = [newPopList begin: [self getZone]];

    while ( ( member = [index next]) ) {
      char* saux = calloc( STRSIZE, sizeof( char ) );
                                // Must be allocated to be used as key
      float result =  ((*evalFunc)( member )) ;
      [member setFitness: result];
      sprintf( saux, "%f", result );

      // test to see a member's fitness
      printf("the member's fitness is: %f at generation %d\n", result, 
generationCount);       

      if (! [popList at: (id) saux insert: member ] ) {
        [SourceMessage raiseEvent : "Something is wrong\n" ];
      }
    }
    [index drop];

    // now remove all members of the list
    index = [newPopList begin: [self getZone]];
    while ( ( member = [index next]) ) {
        [index remove];
    }
    [index drop];

    return self;
} 
-----------------------------------------------------------------

And you can see there are still members with fitness values close to 
0.0. Because F5 have many sharp peaks as well as low valleys, this is 
expected. However these lower fitness values are not shown up in the 
minimum fitness sequence in the "Fitness Evolution" graph.

Can anyone suggest what has gone wrong here?

Thanks for the help.

-- Xiaodong
 



                  ==================================
   Swarm-Modelling is for discussion of Simulation and Modelling techniques
   esp. using Swarm.  For list administration needs (esp. [un]subscribing),
   please send a message to <address@hidden> with "help" in the
   body of the message.
                  ==================================