Parsing unsorted data from large fixed width text - perl

Question

I am mostly a Matlab user and a Perl n00b. This is my first Perl script.
I have a large fixed width data file that I would like to process into a binary file with a table of contents. My issue is that the data files are pretty large and the data parameters are sorted by time. Which makes it difficult (at least for me) to parse into Matlab. So seeing how Matlab is not that good at parsing text I thought I would try Perl. I wrote the following code which works ... at least on my small test file. However it is painfully slow when I tried it on an actual large data file. It was pieced together which lots of examples for various tasks from the web / Perl documentation.
Here is a small sample of the data file. Note: Real file has about 2000 parameter and is 1-2GB. Parameters can be text, doubles, or unsigned integers.
Param 1 filter = ALL_VALUES
Param 2 filter = ALL_VALUES
Param 3 filter = ALL_VALUES
Time Name Ty Value
---------- ---------------------- --- ------------
1.1 Param 1 UI 5
2.23 Param 3 TXT Some Text 1
3.2 Param 1 UI 10
4.5 Param 2 D 2.1234
5.3 Param 1 UI 15
6.121 Param 2 D 3.1234
7.56 Param 3 TXT Some Text 2
The basic logic of my script is to:
Read until the ---- line to build list of parameters to extract (always has "filter =").
Use the --- line to determine field widths. It is broken by spaces.
For each parameter build time and data array (while nested inside of foreach param)
In continue block write time and data to binary file. Then record name, type, and offsets in text table of contents file (used to read the file later into Matlab).
Here is my script:
#!/usr/bin/perl
$lineArg1 = #ARGV[0];
open(INFILE, $lineArg1);
open BINOUT, '>:raw', $lineArg1.".bin";
open TOCOUT, '>', $lineArg1.".toc";
my $line;
my $data_start_pos;
my #param_name;
my #template;
while ($line = <INFILE>) {
chomp $line;
if ($line =~ s/\s+filter = ALL_VALUES//) {
$line = =~ s/^\s+//;
$line =~ s/\s+$//;
push #param_name, $line;
}
elsif ($line =~ /^------/) {
#template = map {'A'.length} $line =~ /(\S+\s*)/g;
$template[-1] = 'A*';
$data_start_pos = tell INFILE;
last; #Reached start of data exit loop
}
}
my $template = "#template";
my #lineData;
my #param_data;
my #param_time;
my $data_type;
foreach $current_param (#param_name) {
#param_time = ();
#param_data = ();
seek(INFILE,$data_start_pos,0); #Jump to data start
while ($line = <INFILE>) {
if($line =~ /$current_param/) {
chomp($line);
#lineData = unpack $template, $line;
push #param_time, #lineData[0];
push #param_data, #lineData[3];
}
} # END WHILE <INFILE>
} #END FOR EACH NAME
continue {
$data_type = #lineData[2];
print TOCOUT $current_param.",".$data_type.",".tell(BINOUT).","; #Write name,type,offset to start time
print BINOUT pack('d*', #param_time); #Write TimeStamps
print TOCOUT tell(BINOUT).","; #offset to end of time/data start
if ($data_type eq "TXT") {
print BINOUT pack 'A*', join("\n",#param_data);
}
elsif ($data_type eq "D") {
print BINOUT pack('d*', #param_data);
}
elsif ($data_type eq "UI") {
print BINOUT pack('L*', #param_data);
}
print TOCOUT tell(BINOUT).","."\n"; #Write memory loc to end data
}
close(INFILE);
close(BINOUT);
close(TOCOUT);
So my questions to you good people of the web are as follows:
What am I obviously screwing up? Syntax, declaring variables when I don't need to, etc.
This is probably slow (guessing) because of the nested loops and searching the line by line over and over again. Is there a better way to restructure the loops to extract multiple lines at once?
Any other speed improvement tips you can give?
Edit: I modified the example text file to illustrate non-integer time stamps and Param Names may contain spaces.

Answer 1

First, you should always have 'use strict;' and 'use warnings;' pragmas in your script.
It seems like you need a simple array (#param_name) for reference, so loading those values would be straight forward as you have it. (again, adding the above pragmas would start showing you errors, including the $line = =~ s/^\s+//; line!)
I suggest you read this, to understand how you can load your data file into a
Hash of Hashes. Once you've designed the hash, you simply read and load the file data contents, and then iterate through the contents of the hash.
For example, using time as the key for the hash
%HoH = (
1 => {
name => "Param1",
ty => "UI",
value => "5",
},
2 => {
name => "Param3",
ty => "TXT",
value => "Some Text 1",
},
3 => {
name => "Param1",
ty => "UI",
value => "10",
},
);
Make sure you close the INFILE after reading in the contents, before you start processing.
So in the end, you iterate over the hash, and reference the array (instead of the file contents) for your output writes - I would imagine it would be much faster to do this.
Let me know if you need more info.
Note: if you go this route, include Data:Dumper - a significant help to printing and understanding the data in your hash!

Answer 2

It seems to me that embedded spaces can only occur in the last field. That makes using split ' ' feasible for this problem.
I am assuming you are not interested in the header. In addition, I am assuming you want a vector for each parameter and are not interested in timestamps.
To use data file names specified on the command line or piped through standard input, replace <DATA> with <>.
#!/usr/bin/env perl
use strict; use warnings;
my %data;
$_ = <DATA> until /^-+/; # skip header
while (my $line = <DATA>) {
$line =~ s/\s+\z//;
last unless $line =~ /\S/;
my (undef, $param, undef, $value) = split ' ', $line, 4;
push #{ $data{ $param } }, $value;
}
use Data::Dumper;
print Dumper \%data;
__DATA__
Param1 filter = ALL_VALUES
Param2 filter = ALL_VALUES
Param3 filter = ALL_VALUES
Time Name Ty Value
---------- ---------------------- --- ------------
1 Param1 UI 5
2 Param3 TXT Some Text 1
3 Param1 UI 10
4 Param2 D 2.1234
5 Param1 UI 15
6 Param2 D 3.1234
7 Param3 TXT Some Text 2
Output:
$VAR1 = {
'Param2' => [
'2.1234',
'3.1234'
],
'Param1' => [
'5',
'10',
'15'
],
'Param3' => [
'Some Text 1',
'Some Text 2'
]
};

Answer 3

First off, this piece of code causes the input file to be read once for every param. Which is quite in-efficient.
foreach $current_param (#param_name) {
...
seek(INFILE,$data_start_pos,0); #Jump to data start
while ($line = <INFILE>) { ... }
...
}
Also there is very rarely a reason to use a continue block. This is more style / readability, then a real problem.
Now on to make it more performant.
I packed the sections individually, so that I could process a line exactly once. To prevent it from using up tons of RAM, I used File::Temp to store the data until I was ready for it. Then I used File::Copy to append those sections into the binary file.
This is a quick implementation. If I were to add much more to it, I would split it up more than it is now.
#!/usr/bin/perl
use strict;
use warnings;
use File::Temp 'tempfile';
use File::Copy 'copy';
use autodie qw':default copy';
use 5.10.1;
my $input_filename = shift #ARGV;
open my $input, '<', $input_filename;
my #param_names;
my $template = ''; # stop uninitialized warning
my #field_names;
my $field_name_line;
while( <$input> ){
chomp;
next if /^\s*$/;
if( my ($param) = /^\s*(.+?)\s+filter = ALL_VALUES\s*$/ ){
push #param_names, $param;
}elsif( /^[\s-]+$/ ){
my #fields = split /(\s+)/;
my $pos = 0;
for my $field (#fields){
my $length = length $field;
if( substr($field, 0, 1) eq '-' ){
$template .= "\#${pos}A$length ";
}
$pos += $length;
}
last;
}else{
$field_name_line = $_;
}
}
#field_names = unpack $template, $field_name_line;
for( #field_names ){
s(^\s+){};
$_ = lc $_;
$_ = 'type' if substr('type', 0, length $_) eq $_;
}
my %temp_files;
for my $param ( #param_names ){
for(qw'time data'){
my $fh = tempfile 'temp_XXXX', UNLINK => 1;
binmode $fh, ':raw';
$temp_files{$param}{$_} = $fh;
}
}
my %convert = (
TXT => sub{ pack 'A*', join "\n", #_ },
D => sub{ pack 'd*', #_ },
UI => sub{ pack 'L*', #_ },
);
sub print_time{
my($param,$time) = #_;
my $fh = $temp_files{$param}{time};
print {$fh} $convert{D}->($time);
}
sub print_data{
my($param,$format,$data) = #_;
my $fh = $temp_files{$param}{data};
print {$fh} $convert{$format}->($data);
}
my %data_type;
while( my $line = <$input> ){
next if $line =~ /^\s*$/;
my %fields;
#fields{#field_names} = unpack $template, $line;
print_time( #fields{(qw'name time')} );
print_data( #fields{(qw'name type value')} );
$data_type{$fields{name}} //= $fields{type};
}
close $input;
open my $bin, '>:raw', $input_filename.".bin";
open my $toc, '>', $input_filename.".toc";
for my $param( #param_names ){
my $data_fh = $temp_files{$param}{data};
my $time_fh = $temp_files{$param}{time};
seek $data_fh, 0, 0;
seek $time_fh, 0, 0;
my #toc_line = ( $param, $data_type{$param}, 0+sysseek($bin, 0, 1) );
copy( $time_fh, $bin, 8*1024 );
close $time_fh;
push #toc_line, sysseek($bin, 0, 1);
copy( $data_fh, $bin, 8*1024 );
close $data_fh;
push #toc_line, sysseek($bin, 0, 1);
say {$toc} join ',', #toc_line, '';
}
close $bin;
close $toc;

Answer 4

I modified my code to build a Hash as suggested. I have not incorporate the output to binary yet due to time limitations. Plus I need to figure out how to reference the hash to get the data out and pack it into binary. I don't think that part should be to difficult ... hopefully
On an actual data file (~350MB & 2.0 Million lines) the following code takes approximately 3 minutes to build the hash. CPU usage was 100% on 1 of my cores (nill on the other 3) and Perl memory usage topped out at around 325MB ... until it dumped millions of lines to the prompt. However the print Dump will be replaced with a binary pack.
Please let me know if I am making any rookie mistakes.
#!/usr/bin/perl
use strict;
use warnings;
use Data::Dumper;
my $lineArg1 = $ARGV[0];
open(INFILE, $lineArg1);
my $line;
my #param_names;
my #template;
while ($line = <INFILE>) {
chomp $line; #Remove New Line
if ($line =~ s/\s+filter = ALL_VALUES//) { #Find parameters and build a list
push #param_names, trim($line);
}
elsif ($line =~ /^----/) {
#template = map {'A'.length} $line =~ /(\S+\s*)/g; #Make template for unpack
$template[-1] = 'A*';
my $data_start_pos = tell INFILE;
last; #Reached start of data exit loop
}
}
my $size = $#param_names+1;
my #getType = ((1) x $size);
my $template = "#template";
my #lineData;
my %dataHash;
my $lineCount = 0;
while ($line = <INFILE>) {
if ($lineCount % 100000 == 0){
print "On Line: ".$lineCount."\n";
}
if ($line =~ /^\d/) {
chomp($line);
#lineData = unpack $template, $line;
my ($inHeader, $headerIndex) = findStr($lineData[1], #param_names);
if ($inHeader) {
push #{$dataHash{$lineData[1]}{time} }, $lineData[0];
push #{$dataHash{$lineData[1]}{data} }, $lineData[3];
if ($getType[$headerIndex]){ # Things that only need written once
$dataHash{$lineData[1]}{type} = $lineData[2];
$getType[$headerIndex] = 0;
}
}
}
$lineCount ++;
} # END WHILE <INFILE>
close(INFILE);
print Dumper \%dataHash;
#WRITE BINARY FILE and TOC FILE
my %convert = (TXT=>sub{pack 'A*', join "\n", #_}, D=>sub{pack 'd*', #_}, UI=>sub{pack 'L*', #_});
open my $binfile, '>:raw', $lineArg1.'.bin';
open my $tocfile, '>', $lineArg1.'.toc';
for my $param (#param_names){
my $data = $dataHash{$param};
my #toc_line = ($param, $data->{type}, tell $binfile );
print {$binfile} $convert{D}->(#{$data->{time}});
push #toc_line, tell $binfile;
print {$binfile} $convert{$data->{type}}->(#{$data->{data}});
push #toc_line, tell $binfile;
print {$tocfile} join(',',#toc_line,''),"\n";
}
sub trim { #Trim leading and trailing white space
my (#strings) = #_;
foreach my $string (#strings) {
$string =~ s/^\s+//;
$string =~ s/\s+$//;
chomp ($string);
}
return wantarray ? #strings : $strings[0];
} # END SUB
sub findStr { #Return TRUE if string is contained in array.
my $searchStr = shift;
my $i = 0;
foreach ( #_ ) {
if ($_ eq $searchStr){
return (1,$i);
}
$i ++;
}
return (0,-1);
} # END SUB
The output is as follows:
$VAR1 = {
'Param 1' => {
'time' => [
'1.1',
'3.2',
'5.3'
],
'type' => 'UI',
'data' => [
'5',
'10',
'15'
]
},
'Param 2' => {
'time' => [
'4.5',
'6.121'
],
'type' => 'D',
'data' => [
'2.1234',
'3.1234'
]
},
'Param 3' => {
'time' => [
'2.23',
'7.56'
],
'type' => 'TXT',
'data' => [
'Some Text 1',
'Some Text 2'
]
}
};
Here is the output TOC File:
Param 1,UI,0,24,36,
Param 2,D,36,52,68,
Param 3,TXT,68,84,107,
Thanks everyone for their help so far! This is an excellent resource!
EDIT: Added Binary & TOC file writing code.