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

55Re: [genetic_programming] Design Tradeoffs, was Re: Beowulf Clusters

Expand Messages
  • Martin C. Martin
    Oct 2, 2001
    • 0 Attachment
      Sean Luke wrote:
      >
      > On Tuesday, October 2, 2001, at 01:36 PM, Martin C. Martin wrote:
      >
      > But one lesson I took away from that project was
      > that a packed array representation (lil-gp uses 4 bytes per node) is not
      > as easy to hack as I would have liked. Writing custom breeding
      > operators for it is a nontrivial endeavor, especially when you add in
      > special constraints like typing, and special-purpose nodes with unusual
      > needs, like storing arbitrary-length statistics on a per-node basis.
      > 1-byte-per-node arrangements like DGP are IMHO even harder to hack weird
      > stuff into. And as a researcher, that's basically what I am: a
      > weird-stuff-hacker.

      At the risk of straying off topic, I don't see why it should be any
      harder to hack weird stuff. No matter what your representation, you
      need a list somewhere of all possible nodes, their arity, the function
      which implements them, etc. So in both cases you have a table. In the
      "compact" representation, each node in the individual is just an index
      into that array. Depending on how you do the typing, you can put the
      type information in the array. That worked fine for my thesis.

      If you want extra info per instance of node, rather than per type of
      node, you may need to put it in the individual as well, but that
      shouldn't be a big deal if you're using object oriented standards; just
      change the data type you use for Node from "byte" to "struct."

      This seems straightforward to me. What am I missing?

      - Martin
    • Show all 19 messages in this topic