mosesdecoder/scripts/training/convert-moses-ini-to-v2.perl

204 lines
6.3 KiB
Perl
Executable File

#!/usr/bin/perl -w
use strict;
my $header = "";
my @INI = <STDIN>;
my %TTABLE_IMPLEMENTATION = ( 0 => "PhraseDictionaryMemory",
1 => "PhraseDictionaryBinary" ,
6 => "PhraseDictionaryMemory");
my %LM_IMPLEMENTATION = ( 0 => "SRILM",
8 => "KENLM lazyken=0" );
my (%FEATURE,%WEIGHT);
my $i=0;
my ($has_sparse_ttable_features,$sparse_weight_file) = (0);
for(; $i<scalar(@INI); $i++) {
my $line = $INI[$i];
if ($line =~ /^\[(.+)\]/) {
my $section = $1;
if ($section eq "ttable-file" ||
$section eq "distortion-file" ||
$section eq "generation-file" ||
$section eq "lmodel-file" ||
$section eq "ttable-limit" ||
$section eq "target-word-insertion-feature" ||
$section eq "source-word-deletion-feature" ||
$section eq "word-translation-feature" ||
$section eq "phrase-length-feature") {
$FEATURE{$section} = &get_data();
}
elsif ($section eq "weight-file") {
print $header.$line;
my $WEIGHT_FILE = &get_data();
$sparse_weight_file = $$WEIGHT_FILE[0];
$has_sparse_ttable_features = `cat $sparse_weight_file | grep ^stm | wc -l`;
if ($has_sparse_ttable_features) {
print STDERR "sparse weight feature file has translaton model features\n -> creating new sparse weight file '$sparse_weight_file.new'\n";
print "$sparse_weight_file.new\n";
}
else {
print "$sparse_weight_file\n";
}
}
elsif ($section =~ /weight-(.+)/ && $section ne "weight-file") {
$WEIGHT{$1} = &get_data();
}
elsif ($section eq "report-sparse-features") {
&get_data(); # ignore
}
else {
print STDERR "include section [$section] verbatim.\n";
print $header.$line;
my $SECTION = &get_data();
foreach (@{$SECTION}) {
print $_."\n";
}
}
$header = "";
}
else {
$header .= $line;
}
}
print $header;
if ($has_sparse_ttable_features) {
open(SPARSE,$sparse_weight_file);
open(NEW,">$sparse_weight_file.new");
while(<SPARSE>) {
if (!/^stm/) {
print NEW $_;
}
else {
s/^stm//;
for (my $i=0;$i<scalar@{$FEATURE{"ttable-file"}};$i++) {
print NEW "TranslationModel$i$_";
}
}
}
close(NEW);
close(SPARSE);
}
my ($feature,$weight) = ("","");
$feature .= "UnknownWordPenalty\n";
$weight .= "UnknownWordPenalty0= 1\n";
$feature .= "WordPenalty\n";
$weight .= "WordPenalty0= ".$WEIGHT{"w"}[0]."\n";
$feature .= "Distortion\n";
$weight .= "Distortion0= ".$WEIGHT{"d"}[0]."\n";
foreach my $section (keys %FEATURE) {
if ($section eq "phrase-length-feature") {
$feature .= "PhraseLengthFeature name=pl\n";
}
elsif ($section eq "target-word-insertion-feature") {
my ($factor,$file) = split(/ /,$FEATURE{$section}[0]);
$feature .= "TargetWordInsertionFeature name=twi factor=$factor";
$feature .= " path=$file" if defined($file);
$feature .= "\n";
}
elsif ($section eq "source-word-deletion-feature") {
my ($factor,$file) = split(/ /,$FEATURE{$section}[0]);
$feature .= "SourceWordDeletionFeature name=swd factor=$factor";
$feature .= " path=$file" if defined($file);
$feature .= "\n";
}
elsif ($section eq "word-translation-feature") {
my ($factors,$simple,$dummy1,$dummy2,$dummy3,$dummy4,$file_f,$file_e) = split(/ /,$FEATURE{$section}[0]);
my ($input_factor,$output_factor) = split(/\-/, $factors);
$feature .= "WordTranslationFeature name=wt input-factor=$input_factor output-factor=$output_factor simple=$simple source-context=0 target-context=0";
$feature .= " source-path=$file_f target-path=$file_e" if defined($file_f);
$feature .= "\n";
}
elsif ($section eq "ttable-file") {
my $i = 0;
my @TTABLE_LIMIT = @{$FEATURE{"ttable-limit"}};
my @W = @{$WEIGHT{"t"}};
foreach my $line (@{$FEATURE{$section}}) {
my ($imp, $input_factor, $output_factor, $weight_count, $file) = split(/ /,$line);
my $implementation = $TTABLE_IMPLEMENTATION{$imp};
if (!defined($implementation)) {
print STDERR "ERROR: Unknown translation table implementation: $implementation\n";
$implementation = "UNKNOWN";
}
$feature .= "$implementation name=TranslationModel$i num-features=$weight_count path=$file input-factor=$input_factor output-factor=$output_factor";
$feature .= " table-limit=".$TTABLE_LIMIT[$i] if $#TTABLE_LIMIT >= $i;
$feature .= "\n";
$weight .= "TranslationModel$i=".&get_weights(\@W,$weight_count)."\n";
$i++;
}
}
elsif ($section eq "generation-file") {
my $i = 0;
my @W = @{$WEIGHT{"generation"}};
foreach my $line (@{$FEATURE{$section}}) {
my ($input_factor,$output_factor,$weight_count,$file) = split(/ /,$line);
$feature .= "Generation name=GenerationModel$i num-features=$weight_count path=$file input-factor=$input_factor output-factor=$output_factor\n";
$weight .= "GenerationModel$i=".&get_weights(\@W,$weight_count)."\n";
$i++;
}
}
elsif ($section eq "distortion-file") {
my $i = 0;
my @W = @{$WEIGHT{"d"}};
my $ignore = shift @W;
foreach my $line (@{$FEATURE{$section}}) {
my ($factors,$type,$weight_count,$file) = split(/ /,$line);
my ($input_factor,$output_factor) = split(/\-/, $factors);
$feature .= "LexicalReordering name=LexicalReordering$i num-features=$weight_count type=$type input-factor=$input_factor output-factor=$output_factor path=$file\n";
$weight .= "LexicalReordering$i=".&get_weights(\@W,$weight_count)."\n";
$i++;
}
}
elsif ($section eq "lmodel-file") {
my $i = 0;
my @W = @{$WEIGHT{"l"}};
foreach my $line (@{$FEATURE{$section}}) {
my ($imp,$factor,$order,$file) = split(/ /,$line);
my $implementation = $LM_IMPLEMENTATION{$imp};
if (!defined($implementation)) {
print STDERR "ERROR: Unknown language model implementation: $implementation\n";
$implementation = "UNKNOWN";
}
$feature .= "$implementation name=LM$i factor=$factor path=$file order=$order\n";
$weight .= "LM$i=".&get_weights(\@W,1)."\n";
$i++;
}
}
}
print "\n[feature]\n$feature\n";
print "\n[weight]\n$weight\n";
sub get_data {
my ($pattern) = @_;
my @DATA;
while (++$i < scalar(@INI) &&
$INI[$i] !~ /^\s*$/ &&
$INI[$i] !~ /^\[/ &&
$INI[$i] !~ /^\#/) {
push @DATA,$INI[$i];
}
$i--;
chop(@DATA);
return \@DATA;
}
sub get_weights {
my ($W,$count) = @_;
my $list = "";
for(my $w=0;$w<$count;$w++) {
my $value = shift @{$W};
$list .= " $value";
}
return $list;
}