Comments on: Genetic Algorithms in Perl

By: JJ

JJ — Fri, 12 Dec 2008 16:26:38 +0000

Actually, Perl is not so slow for the bread and butter of GA, string handling. And, of course, you (and mostly everybody else) is able to write faster and tighter code in a language they know well (which, in my case, is Perl), than in other that’s supposed to be faster but is unknown.
BTW, you can use anything as a genome in Algorithm::Evolutionary, as long as you match the operators you use with it. There are even several underlying representations for things like bitstrings (string vs. bit vector, guess which one is faster?)

By: sethjust

sethjust — Fri, 12 Dec 2008 08:21:43 +0000

I’ve been playing with AI::Genetic and it’s quite good… I especially value the ability to define your own strategies, as well as the flexibility in defining the genome.

As to performance, I wouldn’t be using Perl if I wanted it to be fast, only if I wanted it to be pretty, which this is. Perl makes for a nice interface to a genetic algorithm, and makes it very easy to modify to any problem. Even my homegrown code has that advantage, and it’s even more present in the modules that are available.

By: JJ

JJ — Fri, 12 Dec 2008 08:16:07 +0000

I hadn’t seen AI::Genetic::Pro; it looks nice, and it’s been XS-ed for performance. However, it doesn’t look too flexible. I did Algorithm::Evolutionary for my own research, and thus what I valued the most was the possibility to easily add new stuff.

By: JJ

JJ — Fri, 12 Dec 2008 08:13:52 +0000

WRT to performance, you can get strong gains by compiling your own perl, eliminating threads and some other features.

By: sethjust

sethjust — Fri, 12 Dec 2008 00:11:33 +0000

I would definitely use a module as well, over home-grown code, both for performance and reliability. However, it’s a fun learning experience. And you’re right, there is some inefficiency in doing the straight sort(), so it’s good to learn a bit about better ways of doing the same thing.

By: Ivan

Ivan — Thu, 11 Dec 2008 23:42:44 +0000

as an alternative to Algo::Evolutionary, http://search.cpan.org/search?query=genetic&mode=all shows also AI::Genetic and AI::Genetic::Pro. I know you did this for fun, but if I was to use Perl genetic algos for serious work, I’d probably go for the ::Pro module.

Another comment on your code – sort {fitness($b) fitness($a)} – will evaluate your fitness function ~ N log(N) times, which is usually not what you want. For education, you should probably read about Orcish maneuver and Schwartzian Transform… as for real work you should probably use Sort::Key.

By: sethjust

sethjust — Thu, 11 Dec 2008 17:09:40 +0000

Thanks for the replies…

@JJ: I hadn’t seen that module. It looks interesting. However, the point here was to write one from scratch, which seems to have worked out fairly well. I wasn’t sure how Perl would do, and it seems to have come through admirably.

@Ivan: Good call on anonymous functions. I’ve made that change and I’ll post updated code in a little bit.

By: Ivan

Ivan — Thu, 11 Dec 2008 15:58:53 +0000

Don’t use string eval – it slows you down significantly, and it’s rather inelegant.

what you are looking for is anonymous functions -

my $init = sub { rand(10) };

and then replace all eval($init_string) with $init->();

By: JJ

JJ — Thu, 11 Dec 2008 13:27:15 +0000

Or… you can sudo cpan Algorithm::Evolutionary and download a nifty and venerable cpan module.
The code looks nice, anyways, and just goes out to prove that Perl is as good a language as any other for programming evolutionary algorithms. Better than many others, in fact.