>I thought initialy an easy way to construct the
>fitnnes function would be:
>Each programm would have a probability of survival. At
>the end of a turn a random mechanism would decide if a
>programm would survive to reproduce or die according
>to the probability..
>If the programmer does not set a probability of
>survival to a programm then we can do lots of things
>like it inherits an average of its parents
>probabilities...
>
>Usualy the fitness function is refered as an important
>part of the "enviroment" where the indiviuals are
>living. We could say that in a "live coding gp system"
>
>(lcgps(hehehe)) the enviroment would be the
>music-programmer! He can edit a given proggram, he can
>kill anyone that's bothering....
Generally speaking, I think combining interactive programming with
genetic programming is interesting because of a certain
incongruousness they seem to have. GP really comes from the tradition
of problem-solving automata, interactive programming more from the
side of expressing problems. If a gp algorithm has found a 'good'
solution, it might not express a problem clearly, it is still a good
solution (and maybe a better one than one could have found by
thinking about it). A question to you who have a certain experience
with this 'breeding' practise: can you still follow the (textual)
output of your algorithms - I mean do they "make sense"? (Of course
in live ecoding it does not matter so much that we, as a programmer,
watch the program program, and, just like the audience, don't
understand what's going on)
Another side of this is that the terminology of genetics (and
immunology) is full of metaphors with a very political taste, usually
neo-liberal: competition, survival, defence etc. This terminology
also naturalizes certain behaviours - like hereditary descendence -
which should nearly automatically be challenged by a process of live
coding. How does literature and nature metaphors "interbreed"?
>
>If a fitness function for scoring music was programmed to score fast melodies
>highly, the selected individuals would favor speed, and the population as a
>whole would be slowly filled with programs that create faster melodies. If you
>could change the fitness function on the fly, your population would react to
>the changing pressures and adapt further to gain high marks.
>
Writing programs on the fly, I got the impression that the relation
between statistical and deteministic properties of the algorithm gets
quite relevant. If there is too much randomness, modifying the
program does not yield any new insight in the process. But this is
the case for a certain durational range - very fast changes are
overlooked easily enough. I would guess that gp + ip is really good
for these ranges.
What do you think?
--
.
Received on Fri Apr 01 2005 - 11:04:56 BST