Finding equal lines in file with Perl - perl

Question

I have a CSV file which contains duplicated items in different rows.
x1,y1
x2,y2
y1,x1
x3,y3
The two rows containing x1,y1 and y1,x1 are a match as they contain the same data in a diffrent order.
I need your help to find an algorithm to search for such lines in a 12MB file.

Answer 1

If you can define some ordering and equality relations between fields, you could store a normalized form and test your lines for equality against that.
As an example, we will use string comparision for your fields, but after lowercasing them. We can then sort the parts according to this relation, and create a lookup table via a nested hash:
use strict; use warnings;
my $cache; # A hash of hashes. Will be autovivified later.
while (<DATA>) {
chomp;
my #fields = split;
# create the normalized representation by lowercasing and sorting the fields
my #normalized_fields = sort map lc, #fields;
# find or create the path in the lookup
my $pointer = \$cache;
$pointer = \${$pointer}->{$_} for #normalized_fields;
# if this is an unknow value, make it known, and output the line
unless (defined $$pointer) {
$$pointer = 1; # set some defined value
print "$_\n"; # emit the unique line
}
}
__DATA__
X1 y1
X2 y2
Y1 x1
X3 y3
In this example I used the scalar 1 as value of the lookup data structure, but in more complex scenarios the original fields or the line number could be stored here. For the sake of the example, I used space-seperated values here, but you could replace the split with a call to Text::CSV or something.
This hash-of-hashes approach has sublinear space complexity, and worst case linear space complexity. The lookup time only depends on the number (and size) of fields in a record, not on the total number of records.
Limitation: All records must have the same number of fields, or some shorter records could be falsely considered “seen”. To circumvent these problems, we can use more complex nodes:
my $pointer = \$cache;
$pointer = \$$pointer->[0]{$_} for #normalized_fields;
unless (defined $$pointer->[1]) {
$$pointer->[1] = 1; ...
}
or introduce a default value for nonexistant field (e.g. the seperator of the original file). Here an example with the NUL character:
my $fields = 3;
...;
die "record too long" if #fields > $fields;
...; # make normalized fields
push #normalized_fields, ("\x00") x ($fields - #normalized_fields);
...; # do the lookup

Answer 2

A lot depends on what you want to know about duplicate lines once they have been found. This program uses a simple hash to list the line numbers of those lines that are equivalent.
use strict;
use warnings;
my %data;
while (<DATA>) {
chomp;
my $key = join ',', sort map lc, split /,/;
push #{$data{$key}}, $.;
}
foreach my $list (values %data) {
next unless #$list > 1;
print "Lines ", join(', ', #$list), " are equivalent\n";
}
__DATA__
x1,y1
x2,y2
y1,x1
x3,y3
output
Lines 1, 3 are equivalent

Answer 3

Make two hash tables A and B
Stream through your input one line at a time
For the first line pair x and y, use each as key and the other as value for both hash tables (e.g., $A->{x} = y; $B->{y} = x;)
For the second and subsequent line pairs, test if the second field's value exists as a key for either A or B — if it does, you have a reverse match — if not, then repeat the addition process from step 3 to add it to the hash tables

Answer 4

To do a version of amon's answer without a hash table, if your data are numerical, you could:
Stream through input line by line, sorting fields one and two by numerical ordering
Pipe result to UNIX sort on first and second fields
Stream through sorted output line by line, checking if current line matches the previous line (reporting a reverse match, if true)
This has the advantage of using less memory than hash tables, but may take more time to process.

Answer 5

amon already provided the answer I would've provided, so please enjoy this bad answer:
#! /usr/bin/perl
use common::sense;
my $re = qr/(?!)/; # always fails
while (<DATA>) {
warn "Found duplicate: $_" if $_ =~ $re;
next unless /^(.*),(.*)$/;
die "Unexpected input at line $.: $_" if "$1$2" =~ tr/,//;
$re = qr/^\Q$2,$1\E$|$re/
}
__DATA__
x1,y1
x2,y2
y1,x1
x3,y3