Re: AI::Genetic gives me strange warnings

When I run the following code:

#!/usr/bin/perl
use strict;
use warnings;
use AI::Genetic;
use Data::Dumper;

$|=1;
my $ga = AI::Genetic->new
  (
   -fitness => \&my_fitness,
   -type => 'rangevector'
  );

$ga->init([ [0,10], [0,10], [0,10], [0,10] ]);
$ga->sortIndividuals($ga->people());
# Note that using this instead of sortPopulation does NOT
# cause the same warnings:
# my @populace = sort {$a->score <=> $b->score} @{$ga->people()};

sub my_fitness {
  my $genes = shift;
  print ref $genes, "\n";

  my @sorted = sort {
    print Dumper($a,$b); sleep 1;
    $genes->[$a] <=> $genes->[$b]
  } 0..3;

  my $score = $sorted[1] + 2*$sorted[2] + 3*$sorted[3];

  print "DEBUG: score was $score\n";
  return $score;
}
[download]

I get

1037% perl genetic.pl
ARRAY
$VAR1 = 0;
$VAR2 = 1;
$VAR1 = 2;
$VAR2 = 3;
$VAR1 = 1;
$VAR2 = 2;
$VAR1 = 1;
$VAR2 = 3;
$VAR1 = 0;
$VAR2 = 3;
DEBUG: score was 7
ARRAY
$VAR1 = undef;
$VAR2 = undef;
Use of uninitialized value in array element at genetic.pl line 27.
Use of uninitialized value in array element at genetic.pl line 27.
$VAR1 = undef;
$VAR2 = undef;
Use of uninitialized value in array element at genetic.pl line 27.
Use of uninitialized value in array element at genetic.pl line 27.
$VAR1 = undef;
$VAR2 = undef;
Use of uninitialized value in array element at genetic.pl line 27.
Use of uninitialized value in array element at genetic.pl line 27.
$VAR1 = undef;
$VAR2 = undef;
Use of uninitialized value in array element at genetic.pl line 27.
Use of uninitialized value in array element at genetic.pl line 27.
DEBUG: score was 14
   etc...
[download]

So the first sort seems to work fine, but it crashes on subsequent iterations. A mystery...

Update: It is an official bug: check out nested sort in perl-5.7.0 fails.

-Mark

Comment on Re: AI::Genetic gives me strange warnings Select or Download Code

Replies are listed 'Best First'.
Re^2: AI::Genetic gives me strange warnings by fizbin (Chaplain) on Feb 23, 2005 at 04:30 UTC
Right - the second time you try that sort, $a and $b are undef each time the block is evaluated. Furthermore, it appears that sort-within-sort only breaks if the inner sort is in a different package than the outer sort, as you'll notice that when I do the sort myself (inside main) there's no problem. Also, see my other reply that exhibits this phenomenon without using AI::Genetic. `-- @/=map{[/./g]}qw/.h_nJ Xapou cets krht ele_ r_ra/; map{y/X_/\n /;print}map{pop@$_}@/for@/` [download]	[reply] [d/l]
Re^3: AI::Genetic gives me strange warnings by Tanktalus (Canon) on Feb 23, 2005 at 15:10 UTC
As ugly as it seems, a workaround (not denying a likely perl bug here) is to use $::a and $::b in the inner sort. Which may mean patching AI::Genetic... although I don't know what that will do during non-embedded sorts.	[reply]
Re^4: AI::Genetic gives me strange warnings by fizbin (Chaplain) on Feb 23, 2005 at 15:23 UTC
Actually, there's a much, much easier patch to AI::Genetic that doesn't potentially screw up other non-embedded sorts. I've already communicated this to the AI::Genetic author, but you can see what's necessary in (the updated) 433606. It relies on the fact that individuals in an AI::Genetic population remember their score values once they've computed them. (Which is necessary for even vaguely decent performance when the fitness function gets a bit hairy) It's a very small three-line change, and does leave open the possibility of people screwing with the internals of the AI::Genetic object if they assign to $_, but you shouldn't be assigning to $_ in any case without using `local` to limit the effects of your changes. Also note that it's not the inner sort you'd be changing if you're changing AI::Genetic but the outer sort; the inner sort is already inside package main. `-- @/=map{[/./g]}qw/.h_nJ Xapou cets krht ele_ r_ra/; map{y/X_/\n /;print}map{pop@$_}@/for@/` [download]	[reply] [d/l] [select]


Perl: the Markov chain saw
	PerlMonks