Re: Generate unique ids of maximum length

[rubasov]

Another idea probably worth mentioning: embrace tab completion. Let's suppose we can use tab completion on your input, for instance consider this line:

Lenoc3_duallayer_1
^^   ^ ^         ^

You must type the marked characters, but the others are optional, they can be completed by pressing tab. Therefore in the shortening process always keep the mandatory characters, but leave out as many optional as needed starting from the right end.

limit  shortened
   10  Lenoc3_du1
       ...
    6  Len3d1
    5  Le3d1
    4  not possible

This is achievable by parsing your input into a suffix tree and operating on that.

L -> XP_ -> 0 -> ...
         -> 1 -> ...
  -> enoc -> 3_ -> carina_ -> ...
                -> duallayer_ -> ...
          -> 5_ -> carina -> ...
                -> duallayer -> ...

The first characters of the tree node strings are the mandatory ones, the others are optional. This tree structure seems similar to ikegami's code, however I haven't understood that fully yet, so I'm not sure what's the difference. For me the benefit of this approach seems to be this: you don't need to use the concept of word. Another pro for this algorithm is that you don't need to number your shortened identifiers. (But numbering your input lines (in the base of your input character set) is not bad because that produces the shortest possible unique ids.)

I have some proof of concept code, but take it with a grain of salt: it's unnecessarily complex, employs dirty hacks etc.

use strict;
use warnings;
use Data::Dump qw( dd );

sub _collapse {
  my $tree = shift;
  my ( $stree, $append );

  if ( ref $tree ) {
    my @keys = keys %$tree;
    if ( @keys == 1 and $keys[0] ne '' ) {
      ( $stree, $append ) = _collapse( $tree->{ $keys[0] } );
      return $stree, defined $append ? $keys[0] . $append : $keys[0];
    }
    else {
      for (@keys) {
        ( my $ref, $append ) = _collapse( $tree->{$_} );
        $stree->{ defined $append ? $_ . $append : $_ } = $ref;
      }
      return $stree;
    }
  }
  return;
}

sub collapse {
  my $ctree = shift;
  my ( $stree, $append ) = _collapse($ctree);
  if ( defined $append ) {
    return { $append => $stree };
  }
  else {
    return $stree;
  }
}

sub shorten {
  my $stree = shift;
  my $limit = shift;

  if ( ref $stree ) {
    while ( my ( $k, $v ) = each %$stree ) {
      local our @parts = @parts;
      push @parts, $k if $k ne '';
      if ( $k eq '' ) {
        if ( @parts > $limit ) {
          print "!\n";
          next;
        }
        my $remaining = $limit - @parts;
        my $shortened = '';
        for ( 0 .. $#parts ) {
          $shortened .= substr $parts[$_], 0, 1;
          my $str = substr $parts[$_], 1, $remaining;
          $shortened .= $str;
          last if ( ( $remaining -= length $str ) < 0 );
        }
        print $shortened, "\t", join( '', @parts ), "\n";
      }
      shorten( $v, $limit );
    }
  }
}

my $ctree = {};

while (<DATA>) {
  chomp;
  my $ref = $ctree;
  for ( split // ) {
    no warnings 'void';
    $ref->{$_}->{''}; # looks like a decent autovivification bug ;-)
    $ref = $ref->{$_};
  }
  $ref->{''} = undef;
}

#dd $ctree;
my $stree = collapse($ctree);
#dd $stree;
shorten( $stree => 5 );

__DATA__
A2990_duallayer_1
A2990_duallayer_2
A2990_duallayer_3
A2990_duallayer_4
A2990_duallayer_5
A2990_duallayer_6
A2990_duallayer_7
A2990_duallayer_8
A2990_duallayer_9
A2990_duallayer_10
LXP_01
LXP_02
LXP_03
LXP_04
LXP_05
LXP_06
LXP_07
LXP_08
LXP_09
LXP_10
LXP_11
LXP_12
LXP_13
LXP_14
LXP_15
LXP_16
LXP_17
LXP_18
Normal_1
Normal_2
Normal_3
Normal_4
Normal_5
Normal_6
Lenoc3_carina_A
Lenoc3_carina_B
Lenoc3_carina_C
Lenoc3_duallayer_1
Lenoc3_duallayer_2
Lenoc3_duallayer_3
Lenoc5_carina_1
Lenoc5_carina_2
Lenoc5_carina_3
Lenoc5_duallayer_1
Lenoc5_duallayer_2
Lenoc5_duallayer_3

Some explanation: I've started by parsing the input into a character-level suffix tree. This tree uses hashes, the keys are the substrings, the values are hashrefs of what can follow. A possible ending position of the string is marked by "" => undef. Then collapsed it to a substring-level suffix tree and descended into that to shorten the ids.

Sorry for the late post, yesterday I haven't got enough time to write this.

p.s.: Am I right this is guaranteed to produce unique ids? I'm not sure, but it seems good.

update: answered my own question: Yes, the uniqueness of the shortened ids is guaranteed, no matter what set of optional characters are left out.