A quick search on CPAN yielded no useful results, though Kullback Leibler Perl yielded a few results.

If you've already gone through the trouble of coding, it might be worth while to post the code here for some aid in optimization. Have you run it through Devel::NYTProf to see where your bottle necks are?

You may use a shell script, that calls perl (do the computations in perl) and then call R. R has a FNN package/module which has Kl.dist function which you may use for K-L distance.

Update: I have written the KL Divergence for discrete probability distribution samples. It is very naive but might be useful.

#!/usr/bin/perl -w

## for Discrete Probability Distribution
## considering that sum of probability values in both arrays equals to
+ 1
 
my $dist = 0;
my @terms = ('a', 'b','c','d');
my @P = (0, 0.3, 0.4, 0.3);
my @Q = (0.2, 0.2, 0.3, 0.3);
        
if ( (scalar(@P) != scalar(@terms) ) && (scalar(@Q) != scalar(@terms))
+ ){
    print " The size should be same \n";
    exit;
}
else{
    for(my $i = 0; $i<= $#P; $i++){
        my $temp = 0 if($P[$i] == 0 || $Q[$i] == 0);
        $temp = $P[$i]*log($P[$i]/$Q[$i]) if($P[$i] != 0 && $Q[$i] != 
+0); 
        $dist = $dist + $temp;
    }
}
print "The Kullback Distance symmetric for discrete Distribution is :"
+, $dist,"\n";
[download]