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

Inheriting genes from the less fit parent during crossover

Expand Messages
  • aimike002
    As I understand NEAT, during crossover the fitter parent genes predominate, or if equal fitness, then inheritance is at random. I found a thread that discusses
    Message 1 of 7 , Jun 2, 2007
    • 0 Attachment
      As I understand NEAT, during crossover the fitter parent genes
      predominate, or if equal fitness, then inheritance is at random.

      I found a thread that discusses inheriting genes from the less fit
      parent during crossover:

      http://tech.groups.yahoo.com/group/neat/message/1443

      I'm thinking about writing my code to do crossover so that genes
      (matching and non matching) are inherited
      at random in proportion to the relative fitness of each parent.

      Maybe the answer is try it and see, but I would be interested if
      anyone has thoughts on this or has tried something
      else along these lines already.

      Mike
    • Kenneth Stanley
      Mike, My guess is that it would work fine, but I m uncertain if you will notice a performance increase. When I first wrote NEAT, I let them inherit all genes
      Message 2 of 7 , Jun 2, 2007
      • 0 Attachment
        Mike,

        My guess is that it would work fine, but I'm uncertain if you will
        notice a performance increase. When I first wrote NEAT, I let them
        inherit all genes from both parents and while it did cause genomes to
        grow faster, the overall performance didn't seem that bad even then.
        In any case, your suggestion sounds well-principled so in that regard
        it's probably a nice way to do it.

        ken

        --- In neat@yahoogroups.com, "aimike002" <aimike002@...> wrote:
        >
        > As I understand NEAT, during crossover the fitter parent genes
        > predominate, or if equal fitness, then inheritance is at random.
        >
        > I found a thread that discusses inheriting genes from the less fit
        > parent during crossover:
        >
        > http://tech.groups.yahoo.com/group/neat/message/1443
        >
        > I'm thinking about writing my code to do crossover so that genes
        > (matching and non matching) are inherited
        > at random in proportion to the relative fitness of each parent.
        >
        > Maybe the answer is try it and see, but I would be interested if
        > anyone has thoughts on this or has tried something
        > else along these lines already.
        >
        > Mike
        >
      • aimike002
        Ken, I ve coded this as a user option. Thanks for the advice. Mike
        Message 3 of 7 , Jun 3, 2007
        • 0 Attachment
          Ken,

          I've coded this as a user option.

          Thanks for the advice.

          Mike



          --- In neat@yahoogroups.com, "Kenneth Stanley" <kstanley@...> wrote:
          >
          > Mike,
          >
          > My guess is that it would work fine, but I'm uncertain if you will
          > notice a performance increase. When I first wrote NEAT, I let them
          > inherit all genes from both parents and while it did cause genomes to
          > grow faster, the overall performance didn't seem that bad even then.
          > In any case, your suggestion sounds well-principled so in that regard
          > it's probably a nice way to do it.
          >
          > ken
          >
        • Stephen Waits
          ... So how does it work Mike? Thanks! -- Stephen Waits steve@waits.net http://swaits.com/
          Message 4 of 7 , Jun 3, 2007
          • 0 Attachment

            On Jun 3, 2007, at 12:26 AM, aimike002 wrote:

            I've coded this as a user option.

            So how does it work Mike?

            Thanks!


          • aimike002
            Steve, If we suppose we have parent genes dad & mum and dad is three times as fit as mum, then 2 things are different to the regular approach: 1) for
            Message 5 of 7 , Jun 4, 2007
            • 0 Attachment
              Steve,

              If we suppose we have parent genes 'dad' & 'mum' and dad is three
              times as fit as mum, then 2 things are different to the regular approach:
              1) for matching genes theres a 75% chance of dads weights being passed
              on & a 25% chance of mums being passed on.
              2) For disjoint / excess genes , dads genes will still be passed on,
              with a 25% chance that mums will be included.

              (2) is slightly different from what I said in my OP, but I hope it
              captures the general intention from Colin's post.

              I've various other thoughts on how to implement the general idea, but
              found this the most straightforward.

              Mike


              --- In neat@yahoogroups.com, Stephen Waits <steve@...> wrote:
              >
              >
              > On Jun 3, 2007, at 12:26 AM, aimike002 wrote:
              >
              > > I've coded this as a user option.
              >
              > So how does it work Mike?
              >
              > Thanks!
              >
              > --
              > Stephen Waits
              > steve@...
              > http://swaits.com/
              >
            • Stephen Waits
              ... Ahh yes, thanks for the update on the algorithm; however, I was wondering about how it worked in practice? Thanks, Steve
              Message 6 of 7 , Jun 4, 2007
              • 0 Attachment
                aimike002 wrote:
                >
                > If we suppose we have parent genes 'dad' & 'mum' and dad is three
                > times as fit as mum, then 2 things are different to the regular approach:

                Ahh yes, thanks for the update on the algorithm; however, I was
                wondering about "how it worked" in practice?

                Thanks,
                Steve
              • aimike002
                Sorry , ...of course! I m still at test/debug for the overall program (ie trying to get xor woring), those trial runs are giving me mum/dad genes with similar
                Message 7 of 7 , Jun 4, 2007
                • 0 Attachment
                  Sorry , ...of course!

                  I'm still at test/debug for the overall program (ie trying to get xor
                  woring), those trial runs are giving me mum/dad genes with similar
                  fitnesses, so nothing to report at present.

                  Mike




                  --- In neat@yahoogroups.com, Stephen Waits <steve@...> wrote:
                  >
                  > aimike002 wrote:
                  > >
                  > > If we suppose we have parent genes 'dad' & 'mum' and dad is three
                  > > times as fit as mum, then 2 things are different to the regular
                  approach:
                  >
                  > Ahh yes, thanks for the update on the algorithm; however, I was
                  > wondering about "how it worked" in practice?
                  >
                  > Thanks,
                  > Steve
                  >
                Your message has been successfully submitted and would be delivered to recipients shortly.