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

Re: [neat] Evolution of a Spaceship

Expand Messages
  • Mattias Fagerlund
    Ken, that s great - I included a link to your page on the GeneticArt page. My site s up again - a simple power outage left my web site down for my entire two
    Message 1 of 10 , Feb 2, 2005
      Ken, that's great - I included a link to your page on the GeneticArt page.
      My site's up again - a simple power outage left my web site down for my
      entire two week vacation. First power outage in years (literally, can't
      remember the last time), oh well ;)

      cheers,
      mattias
    • Darren Izzard
      Ken s adventures with the spaceship led me to have a go with the Genetic Art program, and I ve successfully evolved a rather nice shootemup-style alien ship.
      Message 2 of 10 , Feb 2, 2005
        Ken's adventures with the spaceship led me to have a go with the Genetic
        Art program, and I've successfully evolved a rather nice shootemup-style
        alien ship. Quite an addictive little program! It seems to go more in
        the direction you want it to, than other genetic art programs I've seen.

        --Darren
      • Kenneth Stanley
        Hi Darren, can we see your ship? :) I think the fact that it seems to go more in the direction you want it to is significant. For example, BioMorphs are
        Message 3 of 10 , Feb 3, 2005
          Hi Darren, can we see your ship? :)

          I think the fact that it "seems to go more in the direction you want
          it to" is significant. For example, "BioMorphs" are similar:

          http://alife.fusebox.com/morph_lab.html

          But even though at first BioMorphs seem like you will be able to get
          almost anything, you eventually hit a kind of wall where you start
          noticing the limits of the space you're exploring. The big
          difference in the NEAT-based Art tool is that there are no limits to
          the space because it is complexifying. So it is ultimately
          theoretically searching through umlimited dimensions. I don't know
          if there has been an art or breeding tool like that before.

          ken

          --- In neat@yahoogroups.com, Darren Izzard <zysyshelp@a...> wrote:
          > Ken's adventures with the spaceship led me to have a go with the
          Genetic
          > Art program, and I've successfully evolved a rather nice shootemup-
          style
          > alien ship. Quite an addictive little program! It seems to go more
          in
          > the direction you want it to, than other genetic art programs I've
          seen.
          >
          > --Darren
        • Darren Izzard
          Hi Ken - ... Yes, it s in the group s photo album, called Darren s Alien Ship. --Darren
          Message 4 of 10 , Feb 3, 2005
            Hi Ken -

            >can we see your ship? :)

            Yes, it's in the group's photo album, called "Darren's Alien Ship."

            --Darren
          • Mattias Fagerlund
            ... Well, the genetic programming genetic art tools have the same complexification ability, though the evolutionary steps are typically more destructive in my
            Message 5 of 10 , Feb 3, 2005
              > I don't know
              > if there has been an art or breeding tool like that before.

              Well, the genetic programming genetic art tools have the same
              complexification ability, though the evolutionary steps are typically
              more destructive in my experience.

              I've been wracking (sp?) my brain for methods of evolving color shapes
              with neat, but I think I just had an idea - what about a new input
              that's -1 for red, 0 for green and +1 for blue - and let the nn use
              those decide what effect they should have on the output...

              cheers,
              mattias
            • Kenneth Stanley
              ... shapes ... I don t think I completely understand how this idea would work. Where would the inputs be coming from? Aren t colors determined at the outputs
              Message 6 of 10 , Feb 3, 2005
                --- In neat@yahoogroups.com, "Mattias Fagerlund" <mattias@c...> wrote:

                >
                > I've been wracking (sp?) my brain for methods of evolving color
                shapes
                > with neat, but I think I just had an idea - what about a new input
                > that's -1 for red, 0 for green and +1 for blue - and let the nn use
                > those decide what effect they should have on the output...
                >

                I don't think I completely understand how this idea would work. Where
                would the inputs be coming from? Aren't colors determined at the
                outputs and not the inputs?

                This is unrelated but you mentioned before the problem with RGB being
                virtually separate pitures on top of each other, but have you tried
                starting out with all 3 outputs fully connected to each other? That
                way they would have some evolvable influence over each other from the
                start.

                ken
              • Mattias Fagerlund
                Hey, ... The inputs tell the NN which color the output will be interpreted as. I changed it around so that I used three input nodes, r g and b. r was set when
                Message 7 of 10 , Feb 4, 2005
                  Hey,

                  > I don't think I completely understand how this idea would work. Where
                  > would the inputs be coming from? Aren't colors determined at the
                  > outputs and not the inputs?

                  The inputs tell the NN which color the output will be interpreted as. I
                  changed it around so that I used three input nodes, r g and b. r was set
                  when I was calculating the red intensity. g was set when I was
                  calculating the green intensity etc. For each intesity, I flushed the
                  network and calculated the output. Three such outputs were used to
                  generate the finaly pixel color.

                  It sorta worked, but it became prohibitively slow, since it requires
                  more than three times as much work as the bw version (more than three
                  times because there are three extra input nodes for each network).

                  But the pictures looked far better than the previous color pictures that
                  I generated.

                  > This is unrelated but you mentioned before the problem with RGB being
                  > virtually separate pitures on top of each other, but have you tried
                  > starting out with all 3 outputs fully connected to each other? That
                  > way they would have some evolvable influence over each other from the
                  > start.

                  Yep, I tried that, didn't work much better, but that was a while back. I
                  might have needed some more iterations for the color outputs to
                  stabilize a bit more.

                  cheers,
                  mattias
                • Darren Izzard
                  ... We may have been here before, but have you considered trying another color space, such as HSV etc.? --Darren
                  Message 8 of 10 , Feb 4, 2005
                    >The inputs tell the NN which color the output will be interpreted as. I
                    >changed it around so that I used three input nodes, r g and b.
                    >(snip)
                    >But the pictures looked far better than the previous color pictures that
                    >I generated.

                    We may have been here before, but have you considered trying another
                    color space, such as HSV etc.?

                    --Darren
                  • Mattias Fagerlund
                    ... Yep, I tried it since we discussed it last, and it didn t help. I don t know if it would help in this situation either, but I might give it a go. cheers,
                    Message 9 of 10 , Feb 4, 2005
                      > >The inputs tell the NN which color the output will be interpreted as.
                      > I
                      > >changed it around so that I used three input nodes, r g and b.
                      > >(snip)
                      > >But the pictures looked far better than the previous color pictures
                      > that
                      > >I generated.
                      >
                      > We may have been here before, but have you considered trying another
                      > color space, such as HSV etc.?

                      Yep, I tried it since we discussed it last, and it didn't help. I don't
                      know if it would help in this situation either, but I might give it a go.

                      cheers,
                      mattias
                    Your message has been successfully submitted and would be delivered to recipients shortly.