Loading ...
Sorry, an error occurred while loading the content.

Re: [GP] pursers programs

Expand Messages
  • Douglas Mota
    Hi Dean, In my mastering dissertation I ve developed an alternative way to control bloat. In my case, I was evolving assembly programs through linear GP and
    Message 1 of 6 , Mar 1, 2006
    • 0 Attachment
      Hi Dean,
      In my mastering dissertation I've developed an alternative way to control bloat. In my case, I was evolving assembly programs through linear GP and steady-state evolution (no generations).
      This control works this way: when evaluating 2 individuals, the fittest one wins the tournament. But if they have the same fitness the shortest one wins. This tie-breaking promotes the evolution of shorter programs.
      This method has worked very well in my 3 case studies, being able to control the bloat without compromise the fitness' evolutionary performance.
      Since your case is tree GP, you could try to use the tree depth or the number of nodes as the tie-break parameter.
      Follows the link to a book chapter that I've published, where I've used this same bloat control too:
      http://www.springer.com/sgw/cda/frontpage/0,11855,4-175-22-92733168-detailsPage%253Dppmmedia%257CaboutThisBook%257CaboutThisBook,00.html

      Best regards!


      dean_saxton <dean.saxton@...> escreveu:
      I am a final year Civil Engineering Student and am using GPlab to
      search for programs relating sea data to impact pressures.

      My problem is that the results I am getting have very good fitnesses
      but extremely long strings, perhaps a problem of bloat.

      As far as i can I have tried to control bloat etc using dynamic tree
      depths control parameters etc.

      I was wondering if anyone knew of any good purser program to compact
      the size of strings i am getting that is compatible with matlab/gplab.

      Thank you for your time!

      Regards

      Dean Saxton







      SPONSORED LINKS
      Computer science distance education Computer science course Computer science school Computer science degree Computer science and education Computer science education

      ---------------------------------
      YAHOO! GROUPS LINKS


      Visit your group "genetic_programming" on the web.

      To unsubscribe from this group, send an email to:
      genetic_programming-unsubscribe@yahoogroups.com

      Your use of Yahoo! Groups is subject to the Yahoo! Terms of Service.


      ---------------------------------





      Douglas Mota Dias

      ---------------------------------
      Yahoo! Acesso Grátis
      Internet rápida e grátis. Instale o discador agora!

      [Non-text portions of this message have been removed]
    • marmakoide
      Hi I also used this method, to evolve assembly (BlindBuilder, to appear in EUROGP2006), plans for lego-like structures. My fitness measure is made of some
      Message 2 of 6 , Mar 1, 2006
      • 0 Attachment
        Hi
        I also used this method, to evolve assembly (BlindBuilder, to appear
        in EUROGP2006), plans for lego-like structures. My fitness measure is
        made of some objectives to maximize. If the first objective of the indiv
        A is very near from the first objective of indiv B, then I compare A and
        B according to the second objective. The last objective was the size of
        the assembly plan, that must as small as possible. It seems to work
        quite well... But I've also noticed that trick could also fill the
        population (and slow down the evolution :-( ) with just a little set of
        small individuals.



        Le mercredi 01 mars 2006 à 20:41 +0000, Douglas Mota a écrit :
        > Hi Dean,
        > In my mastering dissertation I've developed an alternative way to
        > control bloat. In my case, I was evolving assembly programs through
        > linear GP and steady-state evolution (no generations).
        > This control works this way: when evaluating 2 individuals, the
        > fittest one wins the tournament. But if they have the same fitness the
        > shortest one wins. This tie-breaking promotes the evolution of shorter
        > programs.
        > This method has worked very well in my 3 case studies, being able
        > to control the bloat without compromise the fitness' evolutionary
        > performance.
        > Since your case is tree GP, you could try to use the tree depth or
        > the number of nodes as the tie-break parameter.
        > Follows the link to a book chapter that I've published, where I've
        > used this same bloat control too:
        > http://www.springer.com/sgw/cda/frontpage/0,11855,4-175-22-92733168-detailsPage%253Dppmmedia%257CaboutThisBook%257CaboutThisBook,00.html
        >
        > Best regards!
        >
        >
        > dean_saxton <dean.saxton@...> escreveu:
        > I am a final year Civil Engineering Student and am using GPlab to
        > search for programs relating sea data to impact pressures.
        >
        > My problem is that the results I am getting have very good fitnesses
        > but extremely long strings, perhaps a problem of bloat.
        >
        > As far as i can I have tried to control bloat etc using dynamic tree
        > depths control parameters etc.
        >
        > I was wondering if anyone knew of any good purser program to compact
        > the size of strings i am getting that is compatible with matlab/gplab.
        >
        > Thank you for your time!
        >
        > Regards
        >
        > Dean Saxton
        >
        >
        >
        >
        >
        >
        >
        > SPONSORED LINKS
        > Computer science distance education Computer science course
        > Computer science school Computer science degree Computer science
        > and education Computer science education
        >
        > ---------------------------------
        > YAHOO! GROUPS LINKS
        >
        >
        > Visit your group "genetic_programming" on the web.
        >
        > To unsubscribe from this group, send an email to:
        > genetic_programming-unsubscribe@yahoogroups.com
        >
        > Your use of Yahoo! Groups is subject to the Yahoo! Terms of
        > Service.
        >
        >
        > ---------------------------------
        >
        >
        >
        >
        >
        > Douglas Mota Dias
        >
        > ---------------------------------
        > Yahoo! Acesso Grátis
        > Internet rápida e grátis. Instale o discador agora!
        >
        > [Non-text portions of this message have been removed]
        >
        >
        >
        > SPONSORED LINKS
        > Computer science
        > distance education
        > Computer science
        > course
        > Computer science
        > school
        > Computer science
        > degree
        > Computer science and
        > education
        > Computer science
        > education
        >
        >
        > ______________________________________________________________________
        > YAHOO! GROUPS LINKS
        >
        > 1. Visit your group "genetic_programming" on the web.
        >
        > 2. To unsubscribe from this group, send an email to:
        > genetic_programming-unsubscribe@yahoogroups.com
        >
        > 3. Your use of Yahoo! Groups is subject to the Yahoo! Terms of
        > Service.
        >
        >
        > ______________________________________________________________________
        >






        ___________________________________________________________________________
        Nouveau : téléphonez moins cher avec Yahoo! Messenger ! Découvez les tarifs exceptionnels pour appeler la France et l'international.
        Téléchargez sur http://fr.messenger.yahoo.com
      • Douglas Mota
        Hi, In some preliminary experiments, I ve developed a way to decrease the pressure over GP to evolve small programs. I don t compare the lengths in every
        Message 3 of 6 , Mar 3, 2006
        • 0 Attachment
          Hi,
          In some preliminary experiments, I've developed a way to decrease the pressure over GP to evolve small programs. I don't compare the lengths in every tournament, but with a random frequency. It's defined a parameter (0<param<1) to control the length's pressure. To decide if the lengths will be compared, a number between 0 and 1 is sorted. If this number is lesser than parameter's value, the length's comparison is done. If not, then this comparison is ignored. Thus, this parameter controls the frequency of this comparisons along the complete evolution.
          In my experiments, I've achieved a nice pressure control.

          Regards!


          marmakoide <marmakoide@...> escreveu:
          Hi
          I also used this method, to evolve assembly (BlindBuilder, to appear
          in EUROGP2006), plans for lego-like structures. My fitness measure is
          made of some objectives to maximize. If the first objective of the indiv
          A is very near from the first objective of indiv B, then I compare A and
          B according to the second objective. The last objective was the size of
          the assembly plan, that must as small as possible. It seems to work
          quite well... But I've also noticed that trick could also fill the
          population (and slow down the evolution :-( ) with just a little set of
          small individuals.



          Le mercredi 01 mars 2006 à 20:41 +0000, Douglas Mota a écrit :
          > Hi Dean,
          > In my mastering dissertation I've developed an alternative way to
          > control bloat. In my case, I was evolving assembly programs through
          > linear GP and steady-state evolution (no generations).
          > This control works this way: when evaluating 2 individuals, the
          > fittest one wins the tournament. But if they have the same fitness the
          > shortest one wins. This tie-breaking promotes the evolution of shorter
          > programs.
          > This method has worked very well in my 3 case studies, being able
          > to control the bloat without compromise the fitness' evolutionary
          > performance.
          > Since your case is tree GP, you could try to use the tree depth or
          > the number of nodes as the tie-break parameter.
          > Follows the link to a book chapter that I've published, where I've
          > used this same bloat control too:
          > http://www.springer.com/sgw/cda/frontpage/0,11855,4-175-22-92733168-detailsPage%253Dppmmedia%257CaboutThisBook%257CaboutThisBook,00.html
          >
          > Best regards!
          >
          >
          > dean_saxton <dean.saxton@...> escreveu:
          > I am a final year Civil Engineering Student and am using GPlab to
          > search for programs relating sea data to impact pressures.
          >
          > My problem is that the results I am getting have very good fitnesses
          > but extremely long strings, perhaps a problem of bloat.
          >
          > As far as i can I have tried to control bloat etc using dynamic tree
          > depths control parameters etc.
          >
          > I was wondering if anyone knew of any good purser program to compact
          > the size of strings i am getting that is compatible with matlab/gplab.
          >
          > Thank you for your time!
          >
          > Regards
          >
          > Dean Saxton
          >
          >
          >
          >
          >
          >
          >
          > SPONSORED LINKS
          > Computer science distance education Computer science course
          > Computer science school Computer science degree Computer science
          > and education Computer science education
          >
          > ---------------------------------
          > YAHOO! GROUPS LINKS
          >
          >
          > Visit your group "genetic_programming" on the web.
          >
          > To unsubscribe from this group, send an email to:
          > genetic_programming-unsubscribe@yahoogroups.com
          >
          > Your use of Yahoo! Groups is subject to the Yahoo! Terms of
          > Service.
          >
          >
          > ---------------------------------
          >
          >
          >
          >
          >
          > Douglas Mota Dias
          >
          > ---------------------------------
          > Yahoo! Acesso Grátis
          > Internet rápida e grátis. Instale o discador agora!
          >
          > [Non-text portions of this message have been removed]
          >
          >
          >
          > SPONSORED LINKS
          > Computer science
          > distance education
          > Computer science
          > course
          > Computer science
          > school
          > Computer science
          > degree
          > Computer science and
          > education
          > Computer science
          > education
          >
          >
          > ______________________________________________________________________
          > YAHOO! GROUPS LINKS
          >
          > 1. Visit your group "genetic_programming" on the web.
          >
          > 2. To unsubscribe from this group, send an email to:
          > genetic_programming-unsubscribe@yahoogroups.com
          >
          > 3. Your use of Yahoo! Groups is subject to the Yahoo! Terms of
          > Service.
          >
          >
          > ______________________________________________________________________
          >






          ___________________________________________________________________________
          Nouveau : téléphonez moins cher avec Yahoo! Messenger ! Découvez les tarifs exceptionnels pour appeler la France et l'international.
          Téléchargez sur http://fr.messenger.yahoo.com




          SPONSORED LINKS
          Computer science distance education Computer science course Computer science school Computer science degree Computer science and education Computer science education

          ---------------------------------
          YAHOO! GROUPS LINKS


          Visit your group "genetic_programming" on the web.

          To unsubscribe from this group, send an email to:
          genetic_programming-unsubscribe@yahoogroups.com

          Your use of Yahoo! Groups is subject to the Yahoo! Terms of Service.


          ---------------------------------





          Douglas Mota Dias

          ---------------------------------
          Yahoo! Search
          Dê uma espiadinha e saiba tudo sobre o Big Brother Brasil.

          [Non-text portions of this message have been removed]
        • Sean Luke
          ... This technique is known as Lexicographic Parsimony Pressure (or at least that s what we called it when we did our study on it :-). Here s the relevant
          Message 4 of 6 , Mar 3, 2006
          • 0 Attachment
            Douglas Morta wrote:

            >> In my mastering dissertation I've developed an alternative way to
            >> control bloat. In my case, I was evolving assembly programs through
            >> linear GP and steady-state evolution (no generations).
            >> This control works this way: when evaluating 2 individuals, the
            >> fittest one wins the tournament. But if they have the same fitness
            >> the
            >> shortest one wins. This tie-breaking promotes the evolution of
            >> shorter
            >> programs.

            This technique is known as Lexicographic Parsimony Pressure (or at
            least that's what we called it when we did our study on it :-).
            Here's the relevant paper from GECCO 2002.

            http://cs.gmu.edu/~sean/papers/lexicographic.pdf

            Telated papers:

            Fighting Bloat with Nonparametric Parsimony Pressure (PPSN VII / 2002)
            http://cs.gmu.edu/~sean/papers/parsimony2.pdf

            Alternative Bloat Control Methods (GECCO 2004)
            http://cs.gmu.edu/~lpanait/papers/panait04alternative.pdf

            Sean
          • Douglas Mota
            Hi Sean, Thank you for your explanations and papers references. Since my mastering research that I was looking for a name and some papers that would have used
            Message 5 of 6 , Mar 4, 2006
            • 0 Attachment
              Hi Sean,
              Thank you for your explanations and papers references. Since my mastering research that I was looking for a name and some papers that would have used this same approach already. It's nice know it from you!

              Best Regards!

              Douglas Mota Dias, M.Sc.
              Researcher
              email: douglasm@...-rio.br
              ICA: Applied Computational Intelligence Laboratory
              Electrical Engineering Department
              Pontifícia Universidade Católica do Rio de Janeiro - PUC-Rio
              Brazil
              phone: +55 21 3114-1221
              http://www.ica.ele.puc-rio.br fax: +55 21 3114-1634


              Sean Luke <sean@...> escreveu:
              Douglas Morta wrote:

              >> In my mastering dissertation I've developed an alternative way to
              >> control bloat. In my case, I was evolving assembly programs through
              >> linear GP and steady-state evolution (no generations).
              >> This control works this way: when evaluating 2 individuals, the
              >> fittest one wins the tournament. But if they have the same fitness
              >> the
              >> shortest one wins. This tie-breaking promotes the evolution of
              >> shorter
              >> programs.

              This technique is known as Lexicographic Parsimony Pressure (or at
              least that's what we called it when we did our study on it :-).
              Here's the relevant paper from GECCO 2002.

              http://cs.gmu.edu/~sean/papers/lexicographic.pdf

              Telated papers:

              Fighting Bloat with Nonparametric Parsimony Pressure (PPSN VII / 2002)
              http://cs.gmu.edu/~sean/papers/parsimony2.pdf

              Alternative Bloat Control Methods (GECCO 2004)
              http://cs.gmu.edu/~lpanait/papers/panait04alternative.pdf

              Sean


              SPONSORED LINKS
              Computer science distance education Computer science course Computer science school Computer science degree Computer science and education Computer science education

              ---------------------------------
              YAHOO! GROUPS LINKS


              Visit your group "genetic_programming" on the web.

              To unsubscribe from this group, send an email to:
              genetic_programming-unsubscribe@yahoogroups.com

              Your use of Yahoo! Groups is subject to the Yahoo! Terms of Service.


              ---------------------------------





              ---------------------------------
              Yahoo! Acesso Grátis
              Internet rápida e grátis. Instale o discador agora!

              [Non-text portions of this message have been removed]
            Your message has been successfully submitted and would be delivered to recipients shortly.