I made this thread in old /prog/ but decided to bring it here too What do you guys think of genetic programming? Anyone here has any experience with it?
Name:
Anonymous2013-10-08 22:12
I once tried using a genetic algorithm to make cakes and give them to badly coded AIs. The AIs didn't like any cakes, so I deleted everything in frustration.
>>3 thanks for the links, but I was hoping for some discussion...
Name:
Anonymous2013-10-10 3:48
In my personal experience, I've found that if there's no easier way to solve a problem with Genetic Algorithms, it's always easier to ignore the problem somehow.
Ex: the other week it was brought up at work that we need some way of predicting ``anomalies'' in a plot of {time} x {power spectral density curve pair}, where we have the raw image data available to us, but would rather not use it for technical reasons. This is complicated by the fact that nobody could quite give us an explanation of what counted as an anomaly, and all we had were a set of three hand-generated curves with circles saying `here is an anomaly, look for indications that within X hours you'll see something like this.'
For other various technical reasons, this could be solved by training algorithms on-site, where we would have far more data than three sets of curves, but instead we just added an option to the user interface to show the {time} x {psd} plot until the current moment, plus a big red button for a user to press if they think an anomaly is coming. The customer was satisfied, end of meeting.
I don't mean to imply that Genetic Algorithms aren't useful at all, but perhaps I could state that they are in such a state of research that one who employs GA to solve a problem is likely doing so as much as to explore GA as to solve the problem itself. Then again, perhaps I've been in the Enterprise too long.
GAs can be useful for finding interesting and somewhat efficient solutions to problems that are computationally intractable and/or too complex for a human to handle.