Well, so far, this looks promising. After spending a couple of hours just blindly rushing into it, I'm getting some pretty fast convergence on the basic criteria (rooms as occupied as possible, no room overbooked, no teacher overbooked, rudimentary bias) with the following code. I'm generally seeing convergence in about 20 generations, or about 15 seconds of real time.
#!/usr/bin/perl
use strict;
$|++;
my @CLASSES = split /\s+/, <<'END';
XXX
a-1 a-2 a-3 a-4 a-5
b-1 b-2 b-3 b-4 b-5
c-1 c-2 c-3
d-1 d-2 d-3 d-4 d-5
e-1 e-2 e-3 e-4
f-1 f-2 f-3 f-4 f-5 f-6
g-1
END
my $SLOTS = 6;
my $ROOMS = 3;
use AI::Genetic;
my $ga = AI::Genetic->new
(
-fitness => \&my_fitness,
-type => 'listvector',
-terminate => \&my_terminate,
);
$ga->init([(\@CLASSES) x ($SLOTS * $ROOMS)]);
$ga->evolve(rouletteTwoPoint => 20);
print "final winners\n";
for my $i ($ga->getFittest(5)) {
show_individual($i);
}
sub show_individual {
my $i = shift;
printf "score: %g\n", $i->score;
my @g = $i->genes;
while (@g) {
print " ", join " ", splice @g, 0, $ROOMS;
print "\n";
}
print "\n";
}
sub my_fitness {
my $genes = shift;
my $score = 0;
## process slot by slot
my %seen;
my @g = @$genes;
while (@g) {
my @rooms = splice @g, 0, $ROOMS;
my %teacher_seen;
for (@rooms) {
## "in with the good"...
$score++ if /-/; # good if scheduled (no room left beh
+ind!)
$score += 0.5 if /3/; # good if it's a 3 (simulate user de
+mand)
## "and out with the bad"...
$score -= 100 if /-/ and $seen{$_}++; # bad if duplicated
if (/(.+)-/) { # actual class?
$score -= 100 if $teacher_seen{$1}++;
}
}
}
return $score;
}
sub my_terminate {
my $ga = shift;
print "[", $ga->getFittest->score, "]";
## show_individual($ga->getFittest);
return 0; # do not terminate
}