You have 9 groups with 0-3 numbers in the range 2..9.

I.e each group can be represented with a byte with at most 3 bits set.

There are only 93=56+28+8+1 such combinations possible.

ln(93*9)/ln(256)= 7.35655366 bytes per line

At the moment you'll need -2.5 characters per group which results in -22.5 char per line. (56*3+28*2+8*1)/93

That's about one third.So even with a non binary representation you should achieve your 50 percent or better.

This can only be improved if the combinations don't have the same likelihood.

I don't wanna dig deeper because I don't trust your code and smell an xy problem here.

Update

I just realised that you are forbidding consecutive numbers in your if condition. I.e (2,3,9) is never possible.

This will change the math, but the approach is the same.

Roboticus said you need 15 char in average 7.4 bytes per line is just an upper boundary, so 50% is easily reached.

Don't wanna calculate it again! This would be needed to be done programmatically.

(But I don't trust your code anyway ;)

Instead of calculating I wrote a little script counting the probabilities of group combinations.

Roboticus was right, you need about 14,05 characters per line plus "\n"

Only 1 + 8 + 21 + 20 = 50 combinations are possible per group, resulting in a conservative compression of 50.79 bits = 6.34 bytes per line, which already means a compression to 42% = 58% win.

But if you look into the likelihood of those combinations you clearly see that a Huffman encoding would result in an even better ratio.

I think that's then near the theoretical optimum. (further reading Huffman_coding_with_unequal_letter_costs )

All this supposing your input was real... ;-)

 
use strict;
use warnings;
use Data::Dump qw/pp/;

my %count;

for my $c0 (0..9){
   for my $c1 (0..9){
      for my $c2 (0..9){
         for my $c3 (0..9) {
            my @c = sort {$a <=>$b} ($c0,$c1,$c2,$c3);
            #print "@c\t:\t";
            my @allowed;
            for my $i (1..3) {
               if ( $c[$i] != $c[$i-1]    &&   $c[$i] != $c[$i-1]+1 ) 
+{
                  push @allowed, $c[$i] 
               }
            }
            #print "@allowed\n";
            $count{join "",@allowed}++
         }
      }
   }
}



my @length;
my $average;
for my $k (keys %count){
   my $len = length $k;
   $length[$len]++;
   $average+= $len* $count{$k}/10000;
}

warn '@length: ', pp \@length;
warn 'average #characters: group/line',pp [$average,$average*9];


my  $combies =0;
$combies+= $length[$_] for 0..3;
#$combies=93;

warn "# possible combinations: ", $combies;


my $upper_bound= log($combies)/log(2)*9;
warn 'Upper bound bits, bytes', pp [$upper_bound, $upper_bound/8];

#warn "ranking", pp [ sort {$b <=>$a} values %count ];

warn 'probabilities: ',pp \%count;
[download]

@length: [1, 8, 21, 20] at /tmp/compress.pl line 36.
average #characters: group/line[1.5624, 14.0616] at /tmp/compress.pl l
+ine 37.
# possible combinations: 50 at /tmp/compress.pl line 44.
Upper bound bits, bytes[50.7947057079725, 6.34933821349656] at /tmp/co
+mpress.pl line 48.
probabilities: {
  ""    => 592,
  "2"   => 74,
  "24"  => 60,
  "246" => 24,
  "247" => 24,
  "248" => 24,
  "249" => 24,
  "25"  => 84,
  "257" => 24,
  "258" => 24,
  "259" => 24,
  "26"  => 84,
  "268" => 24,
  "269" => 24,
  "27"  => 84,
  "279" => 24,
  "28"  => 84,
  "29"  => 60,
  "3"   => 208,
  "35"  => 144,
  "357" => 48,
  "358" => 48,
  "359" => 48,
  "36"  => 192,
  "368" => 48,
  "369" => 48,
  "37"  => 192,
  "379" => 48,
  "38"  => 192,
  "39"  => 144,
  "4"   => 366,
  "46"  => 228,
  "468" => 72,
  "469" => 72,
  "47"  => 300,
  "479" => 72,
  "48"  => 300,
  "49"  => 228,
  "5"   => 524,
  "57"  => 312,
  "579" => 96,
  "58"  => 408,
  "59"  => 312,
  "6"   => 682,
  "68"  => 396,
  "69"  => 396,
  "7"   => 840,
  "79"  => 336,
  "8"   => 830,
  "9"   => 508,
} at /tmp/compress.pl line 52.
[download]

Cheers Rolf
_{(addicted to the Perl Programming Language :)
Wikisyntax for the Monastery FootballPerl is like chess, only without the dice}

update

I just realized that Roboticus already had the same basic ideas here: Re: Data compression by 50% + : is it possible?