From e26434979adc33bd949566ba7bf02dff64e80a3e Mon Sep 17 00:00:00 2001 From: Chris Dyer Date: Tue, 2 Oct 2012 00:19:43 -0400 Subject: cdec cleanup, remove bayesian stuff, parsing stuff --- gi/pipeline/OLD.clsp.config | 9 - gi/pipeline/OLD.evaluation-pipeline.pl | 277 -------------- gi/pipeline/backoff-pipe.pl | 215 ----------- gi/pipeline/blacklight.config | 9 - gi/pipeline/clsp.config | 10 - gi/pipeline/evaluation-pipeline.pl | 364 ------------------ gi/pipeline/local-gi-pipeline.pl | 465 ----------------------- gi/pipeline/lticluster.config | 9 - gi/pipeline/scripts/filter-by-f.pl | 56 --- gi/pipeline/scripts/patch-corpus.pl | 65 ---- gi/pipeline/scripts/refilter.pl | 40 -- gi/pipeline/scripts/rekey.pl | 8 - gi/pipeline/scripts/remove-tags-from-contexts.pl | 53 --- gi/pipeline/scripts/remove-tags-from-corpus.pl | 44 --- gi/pipeline/scripts/sort-by-key.sh | 5 - gi/pipeline/scripts/xfeats.pl | 39 -- gi/pipeline/valhalla.config | 9 - 17 files changed, 1677 deletions(-) delete mode 100644 gi/pipeline/OLD.clsp.config delete mode 100755 gi/pipeline/OLD.evaluation-pipeline.pl delete mode 100644 gi/pipeline/backoff-pipe.pl delete mode 100644 gi/pipeline/blacklight.config delete mode 100644 gi/pipeline/clsp.config delete mode 100755 gi/pipeline/evaluation-pipeline.pl delete mode 100755 gi/pipeline/local-gi-pipeline.pl delete mode 100644 gi/pipeline/lticluster.config delete mode 100755 gi/pipeline/scripts/filter-by-f.pl delete mode 100755 gi/pipeline/scripts/patch-corpus.pl delete mode 100755 gi/pipeline/scripts/refilter.pl delete mode 100755 gi/pipeline/scripts/rekey.pl delete mode 100755 gi/pipeline/scripts/remove-tags-from-contexts.pl delete mode 100755 gi/pipeline/scripts/remove-tags-from-corpus.pl delete mode 100755 gi/pipeline/scripts/sort-by-key.sh delete mode 100755 gi/pipeline/scripts/xfeats.pl delete mode 100644 gi/pipeline/valhalla.config (limited to 'gi/pipeline') diff --git a/gi/pipeline/OLD.clsp.config b/gi/pipeline/OLD.clsp.config deleted file mode 100644 index cd0f9d65..00000000 --- a/gi/pipeline/OLD.clsp.config +++ /dev/null @@ -1,9 +0,0 @@ -# THIS FILE GIVES THE LOCATIONS OF THE CORPORA USED -# name path aligned-corpus LM xfeats.grammar dev dev-refs test1 testt-eval.sh ... -btec /export/ws10smt/data/btec/ split.zh-en.al lm/en.3gram.lm.gz xgrammar/grammar.gz devtest/devset1_2.zh devtest/devset1_2.lc.en* devtest/devset3.zh eval-devset3.sh -fbis /export/ws10smt/data/chinese-english.fbis corpus.zh-en.al -zhen /export/ws10smt/data/chinese-english corpus.zh-en.al -aren /export/ws10smt/data/arabic-english corpus.ar-en.al -uren /export/ws10smt/data/urdu-english corpus.ur-en.al -nlfr /export/ws10smt/data/dutch-french corpus.nl-fr.al - diff --git a/gi/pipeline/OLD.evaluation-pipeline.pl b/gi/pipeline/OLD.evaluation-pipeline.pl deleted file mode 100755 index 49c303eb..00000000 --- a/gi/pipeline/OLD.evaluation-pipeline.pl +++ /dev/null @@ -1,277 +0,0 @@ -#!/usr/bin/perl -w -use strict; -use Getopt::Long; -use Cwd; -my $CWD = getcwd; - -my $SCRIPT_DIR; BEGIN { use Cwd qw/ abs_path /; use File::Basename; $SCRIPT_DIR = dirname(abs_path($0)); push @INC, $SCRIPT_DIR; } - -my @DEFAULT_FEATS = qw( - LogRuleCount SingletonRule LexE2F LexF2E WordPenalty - LogFCount LanguageModel Glue GlueTop PassThrough SingletonF -); - -my %init_weights = qw( - LogRuleCount 0.2 - LexE2F -0.3 - LexF2E -0.3 - LogFCount 0.1 - WordPenalty -1.5 - LanguageModel 1.2 - Glue -1.0 - GlueTop 0.00001 - PassThrough -10.0 - SingletonRule -0.1 - X_EGivenF -0.3 - X_FGivenE -0.3 - X_LogECount -1 - X_LogFCount -0.1 - X_LogRuleCount 0.3 - X_SingletonE -0.1 - X_SingletonF -0.1 - X_SingletonRule -0.5 -); - -my $CDEC = "$SCRIPT_DIR/../../decoder/cdec"; -my $PARALLELIZE = "$SCRIPT_DIR/../../vest/parallelize.pl"; -my $EXTOOLS = "$SCRIPT_DIR/../../extools"; -die "Can't find extools: $EXTOOLS" unless -e $EXTOOLS && -d $EXTOOLS; -my $VEST = "$SCRIPT_DIR/../../vest"; -die "Can't find vest: $VEST" unless -e $VEST && -d $VEST; -my $DISTVEST = "$VEST/dist-vest.pl"; -my $FILTSCORE = "$EXTOOLS/filter_score_grammar"; -my $ADDXFEATS = "$SCRIPT_DIR/scripts/xfeats.pl"; -assert_exec($CDEC, $PARALLELIZE, $FILTSCORE, $DISTVEST, $ADDXFEATS); - -my $config = "$SCRIPT_DIR/OLD.clsp.config"; -print STDERR "CORPORA CONFIGURATION: $config\n"; -open CONF, "<$config" or die "Can't read $config: $!"; -my %paths; -my %corpora; -my %lms; -my %devs; -my %devrefs; -my %tests; -my %testevals; -my %xgrammars; -print STDERR " LANGUAGE PAIRS:"; -while() { - chomp; - next if /^#/; - next if /^\s*$/; - s/^\s+//; - s/\s+$//; - my ($name, $path, $corpus, $lm, $xgrammar, $dev, $devref, @xtests) = split /\s+/; - $paths{$name} = $path; - $corpora{$name} = $corpus; - $lms{$name} = $lm; - $xgrammars{$name} = $xgrammar; - $devs{$name} = $dev; - $devrefs{$name} = $devref; - $tests{$name} = $xtests[0]; - $testevals{$name} = $xtests[1]; - print STDERR " $name"; -} -print STDERR "\n"; - -my %langpairs = map { $_ => 1 } qw( btec zhen fbis aren uren nlfr ); - -my $outdir = "$CWD/exp"; -my $help; -my $XFEATS; -my $EXTRA_FILTER = ''; -my $dataDir = '/export/ws10smt/data'; -if (GetOptions( - "data=s" => \$dataDir, - "xfeats" => \$XFEATS, -) == 0 || @ARGV!=2 || $help) { - print_help(); - exit; -} -my $lp = $ARGV[0]; -my $grammar = $ARGV[1]; -print STDERR " CORPUS REPO: $dataDir\n"; -print STDERR " LANGUAGE PAIR: $lp\n"; -die "I don't know about that language pair\n" unless $paths{$lp}; -my $corpdir = "$dataDir"; -if ($paths{$lp} =~ /^\//) { $corpdir = $paths{$lp}; } else { $corpdir .= '/' . $paths{$lp}; } -die "I can't find the corpora directory: $corpdir" unless -d $corpdir; -print STDERR " GRAMMAR: $grammar\n"; -my $LANG_MODEL = mydircat($corpdir, $lms{$lp}); -print STDERR " LM: $LANG_MODEL\n"; -my $CORPUS = mydircat($corpdir, $corpora{$lp}); -die "Can't find corpus: $CORPUS" unless -f $CORPUS; - -my $dev = mydircat($corpdir, $devs{$lp}); -my $drefs = $devrefs{$lp}; -die "Can't find dev: $dev\n" unless -f $dev; -die "Dev refs not set" unless $drefs; -$drefs = mydircat($corpdir, $drefs); - -my $test = mydircat($corpdir, $tests{$lp}); -my $teval = mydircat($corpdir, $testevals{$lp}); -die "Can't find test: $test\n" unless -f $test; -assert_exec($teval); - -if ($XFEATS) { - my $xgram = mydircat($corpdir, $xgrammars{$lp}); - die "Can't find x-grammar: $xgram" unless -f $xgram; - $EXTRA_FILTER = "$ADDXFEATS $xgram |"; - print STDERR "ADDING X-FEATS FROM $xgram\n"; -} - -# MAKE DEV -print STDERR "\nFILTERING FOR dev...\n"; -print STDERR "DEV: $dev (REFS=$drefs)\n"; -`mkdir -p $outdir`; -my $devgrammar = filter($grammar, $dev, 'dev', $outdir); -my $devini = mydircat($outdir, "cdec-dev.ini"); -write_cdec_ini($devini, $devgrammar); - - -# MAKE TEST -print STDERR "\nFILTERING FOR test...\n"; -print STDERR "TEST: $test (EVAL=$teval)\n"; -`mkdir -p $outdir`; -my $testgrammar = filter($grammar, $test, 'test', $outdir); -my $testini = mydircat($outdir, "cdec-test.ini"); -write_cdec_ini($testini, $testgrammar); - - -# CREATE INIT WEIGHTS -print STDERR "\nCREATING INITIAL WEIGHTS FILE: weights.init\n"; -my $weights = mydircat($outdir, "weights.init"); -write_random_weights_file($weights); - - -# VEST -print STDERR "\nMINIMUM ERROR TRAINING\n"; -my $tuned_weights = mydircat($outdir, 'weights.tuned'); -if (-f $tuned_weights) { - print STDERR "TUNED WEIGHTS $tuned_weights EXISTS: REUSING\n"; -} else { - my $cmd = "$DISTVEST --ref-files=$drefs --source-file=$dev --weights $weights $devini"; - print STDERR "MERT COMMAND: $cmd\n"; - `rm -rf $outdir/vest 2> /dev/null`; - chdir $outdir or die "Can't chdir to $outdir: $!"; - $weights = `$cmd`; - die "MERT reported non-zero exit code" unless $? == 0; - chomp $weights; - safesystem($tuned_weights, "cp $weights $tuned_weights"); - print STDERR "TUNED WEIGHTS: $tuned_weights\n"; - die "$tuned_weights is missing!" unless -f $tuned_weights; -} - -# DECODE -print STDERR "\nDECODE TEST SET\n"; -my $decolog = mydircat($outdir, "test-decode.log"); -my $testtrans = mydircat($outdir, "test.trans"); -my $cmd = "cat $test | $PARALLELIZE -j 20 -e $decolog -- $CDEC -c $testini -w $tuned_weights > $testtrans"; -safesystem($testtrans, $cmd) or die "Failed to decode test set!"; - - -# EVALUATE -print STDERR "\nEVALUATE TEST SET\n"; -print STDERR "TEST: $testtrans\n"; -$cmd = "$teval $testtrans"; -safesystem(undef, $cmd) or die "Failed to evaluate!"; -exit 0; - - -sub write_random_weights_file { - my ($file, @extras) = @_; - open F, ">$file" or die "Can't write $file: $!"; - my @feats = (@DEFAULT_FEATS, @extras); - if ($XFEATS) { - my @xfeats = qw( - X_LogRuleCount X_LogECount X_LogFCount X_EGivenF X_FGivenE X_SingletonRule X_SingletonE X_SingletonF - ); - @feats = (@feats, @xfeats); - } - for my $feat (@feats) { - my $r = rand(1.6); - my $w = $init_weights{$feat} * $r; - if ($w == 0) { $w = 0.0001; print STDERR "WARNING: $feat had no initial weight!\n"; } - print F "$feat $w\n"; - } - close F; -} - -sub filter { - my ($grammar, $set, $name, $outdir) = @_; - my $outgrammar = mydircat($outdir, "$name.scfg.gz"); - if (-f $outgrammar) { print STDERR "$outgrammar exists - REUSING!\n"; } else { - my $cmd = "gunzip -c $grammar | $FILTSCORE -c $CORPUS -t $set | $EXTRA_FILTER gzip > $outgrammar"; - safesystem($outgrammar, $cmd) or die "Can't filter and score grammar!"; - } - return $outgrammar; -} - -sub mydircat { - my ($base, $suffix) = @_; - if ($suffix =~ /^\//) { return $suffix; } - my $res = $base . '/' . $suffix; - $res =~ s/\/\//\//g; - return $res; -} - -sub write_cdec_ini { - my ($filename, $grammar_path) = (@_); - open CDECINI, ">$filename" or die "Can't write $filename: $!"; - print CDECINI <> 8; - if ($exitcode) { - print STDERR "Exit code: $exitcode\n"; - if (defined $output && -e $output) { printf STDERR "Removing $output\n"; `rm -rf $output`; } - } - return ! $exitcode; - } -} - -sub assert_exec { - my @files = @_; - for my $file (@files) { - die "Can't find $file - did you run make?\n" unless -e $file; - die "Can't execute $file" unless -e $file; - } -}; - diff --git a/gi/pipeline/backoff-pipe.pl b/gi/pipeline/backoff-pipe.pl deleted file mode 100644 index ac103c8b..00000000 --- a/gi/pipeline/backoff-pipe.pl +++ /dev/null @@ -1,215 +0,0 @@ -#!/usr/bin/perl -w -use strict; - -use Getopt::Long "GetOptions"; - -my @grammars; -my $OUTPUTPREFIX = './giwork/bo.hier.grammar'; -safemkdir($OUTPUTPREFIX); -my $backoff_levels = 1; -my $glue_levels = 1; - -usage() unless &GetOptions('grmr=s@' => \ @grammars, - 'outprefix=s' => \ $OUTPUTPREFIX, - 'bo-lvls=i' => \ $backoff_levels, - 'glue-lvls=i' => \ $glue_levels, -); - -my $OUTDIR = $OUTPUTPREFIX . '/hier'; -print STDERR "@grammars\n"; - - -my %grmr = (); -foreach my $grammar (@grammars) { - $grammar =~ m/\/[^\/]*\.t(\d+)\.[^\/]*/; - $grmr{$1} = $grammar; -} - -my @index = sort keys %grmr; -$OUTDIR = $OUTDIR . join('-',@index); -safemkdir($OUTDIR); -my $BACKOFF_GRMR = $OUTDIR . '/backoff.hier.gz'; -safesystem("echo \"\" | gzip > $BACKOFF_GRMR"); -my $GLUE_GRMR = $OUTDIR . '/glue.hier.gz'; -safesystem("echo \"\" | gzip > $GLUE_GRMR"); -my $joinedgrammars = $OUTDIR . '/grammar.hier.gz'; - -join_grammars(); - -for my $i (0..(scalar @index)-2) { - my $freqs = extract_freqs($index[$i], $index[$i+1]); - if ($i < $backoff_levels) { - create_backoff_rules($index[$i],$index[$i+1],$freqs); - } - if ($i < $glue_levels) { - add_glue_rules($index[$i]); - } -} - -output_grammar_info(); - - -sub usage { - print <> 8; - print STDERR "Exit code: $exitcode\n" if $exitcode; - return ! $exitcode; - } -} - - -sub join_grammars { - print STDERR "\n!!! JOINING GRAMMARS\n"; - if(-e $joinedgrammars) { - print STDERR "$joinedgrammars exists, reusing...\n"; - return; - } - safesystem("echo \"\" | gzip > $joinedgrammars"); - foreach my $i (@index) { - my $g = $grmr{$i}; - safesystem("zcat $g | sed -r -e 's/X([0-9]+)/X$i\\1/g' - | gzip > $g.2.gz"); - safesystem("zcat $joinedgrammars $g.2.gz | gzip > $joinedgrammars.2.gz"); - safesystem("mv $joinedgrammars.2.gz $joinedgrammars"); - } -} - - -sub extract_freqs { - my($grmr1,$grmr2) = @_; - print STDERR "\n!!!EXTRACTING FREQUENCIES: $grmr1->$grmr2\n"; - my $IN_COARSE = substr($grmr{$grmr1},0,index($grmr{$grmr1},".grammar/")) . "/labeled_spans.txt"; - my $IN_FINE = substr($grmr{$grmr2},0,index($grmr{$grmr2},".grammar/")) . "/labeled_spans.txt"; - my $OUT_SPANS = "$OUTDIR/labeled_spans.hier$grmr1-$grmr2.txt"; - my $FREQS = "$OUTDIR/label_freq.hier$grmr1-$grmr2.txt"; - if(-e $OUT_SPANS && -e $FREQS) { - print STDERR "$OUT_SPANS exists, reusing...\n"; - print STDERR "$FREQS exists, reusing...\n"; - return $FREQS; - } - - safesystem("paste -d ' ' $IN_COARSE $IN_FINE > $OUT_SPANS"); - - my %FREQ_HIER = (); - my %finehier = (); - - open SPANS, $OUT_SPANS or die $!; - while () { - my ($tmp, $coarse, $fine) = split /\|\|\|/; - my @coarse_spans = $coarse =~ /\d+-\d+:X(\d+)/g; - my @fine_spans = $fine =~ /\d+-\d+:X(\d+)/g; - - foreach my $i (0..(scalar @coarse_spans)-1) { - my $coarse_cat = $coarse_spans[$i]; - my $fine_cat = $fine_spans[$i]; - - $FREQ_HIER{$coarse_cat}{$fine_cat}++; - } - } - close SPANS; - foreach (values %FREQ_HIER) { - my $coarse_freq = $_; - my $total = 0; - $total+=$_ for (values %{ $coarse_freq }); - $coarse_freq->{$_}=log($coarse_freq->{$_}/$total) for (keys %{ $coarse_freq }); - } - open FREQS, ">", $FREQS or die $!; - foreach my $coarse_cat (keys %FREQ_HIER) { - print FREQS "$coarse_cat |||"; - foreach my $fine_cat (keys %{$FREQ_HIER{$coarse_cat}}) { - my $freq = $FREQ_HIER{$coarse_cat}{$fine_cat}; - print FREQS " $fine_cat:$freq"; - if(! exists $finehier{$fine_cat} || $finehier{$fine_cat} < $freq) { - $finehier{$fine_cat} = $coarse_cat; - } - } - print FREQS "\n"; - } -# foreach my $fine_cat (keys %finehier) { -# print FREQS "$fine_cat -> $finehier{$fine_cat}\n"; -# } - close FREQS; - return $FREQS; -} - - -sub create_backoff_rules { - print STDERR "\n!!! CREATING BACKOFF RULES\n"; - my ($grmr1, $grmr2, $freq) = @_; - my $OUTFILE = "$OUTDIR/backoff.hier$grmr1-$grmr2.txt"; - if(-e $OUTFILE) { - print STDERR "$OUTFILE exists, reusing...\n"; - return; - } - open FREQS, $freq or die $!; - open TMP, ">", $OUTFILE or die $!; - while () { - my $line = $_; - $line = m/^(\d+) \|\|\| (.+)$/; - my $coarse = $1; - $line = $2; - my @finefreq = $line =~ m/(\d+):(\S+)/g; - for(my $i = 0; $i < scalar @finefreq; $i+=2) { - my $finecat = $finefreq[$i]; - my $finefreq = $finefreq[$i+1]; - print TMP "[X$grmr1$coarse] ||| [X$grmr2$finecat,1]\t[1] ||| BackoffRule=$finefreq A=0-0\n"; - } - } - close TMP; - close FREQS; - safesystem("zcat $BACKOFF_GRMR | cat - $OUTFILE | gzip > $BACKOFF_GRMR.2.gz"); - safesystem("mv $BACKOFF_GRMR.2.gz $BACKOFF_GRMR"); -} - -sub add_glue_rules { - print STDERR "\n!!! CREATING GLUE RULES\n"; - my ($grmr) = @_; - my $OUTFILE = "$OUTDIR/glue.$grmr.gz"; - if (-e $OUTFILE) { - print STDERR "$OUTFILE exists, reusing...\n"; - return; - } - open TMP, ">", $OUTFILE or die $!; - for my $i (0..($grmr-1)) { - print TMP "[S] ||| [S,1] [X$grmr$i,2] ||| [1] [2] ||| Glue=1\n"; - print TMP "[S] ||| [X$grmr$i,1] ||| [1] ||| GlueTop=1\n"; - } - close TMP; - safesystem("zcat $GLUE_GRMR | cat - $OUTFILE | gzip > $GLUE_GRMR.2.gz"); - safesystem("mv $GLUE_GRMR.2.gz $GLUE_GRMR"); -} - -sub output_grammar_info { - print STDERR "\n!!! GRAMMAR INFORMATION\n"; - print STDOUT "GRAMMAR: \t$joinedgrammars\n"; - print STDOUT "GLUE: \t$GLUE_GRMR\n"; - print STDOUT "BACKOFF: \t$BACKOFF_GRMR\n"; -} diff --git a/gi/pipeline/blacklight.config b/gi/pipeline/blacklight.config deleted file mode 100644 index fc59a604..00000000 --- a/gi/pipeline/blacklight.config +++ /dev/null @@ -1,9 +0,0 @@ -# THIS FILE GIVES THE LOCATIONS OF THE CORPORA USED -# name path aligned-corpus LM dev dev-refs test1 testt-eval.sh ... -/usr/users/0/cdyer/ws10smt/data -btec /home/cdyer/ws10smt-data/btec/ split.zh-en.al lm/en.3gram.lm.gz devtest/devset1_2.zh devtest/devset1_2.lc.en* devtest/devset3.zh eval-devset3.sh -zhen /home/cdyer/ws10smt-data/chinese-english corpus.zh-en.al lm/c2e.3gram.lm.gz dev_and_test/mt02.src.txt dev_and_test/mt02.ref.* dev_and_test/mt03.src.txt eval-mt03.sh -aren /home/cdyer/ws10smt-data/arabic-english corpus.ar-en-al lm/a2e.3gram.lm.gz dev_and_test/dev.src.txt dev_and_test/dev.ref.txt.* dev_and_test/mt05.src.txt eval-mt05.sh -uren /usr/users/0/cdyer/ws10smt/data/urdu-english corpus.ur-en.al lm/u2e.en.lm.gz dev/dev.ur dev/dev.en* devtest/devtest.ur eval-devtest.sh -nlfr /home/cdyer/ws10smt-data/dutch-french corpus.nl-fr.al - diff --git a/gi/pipeline/clsp.config b/gi/pipeline/clsp.config deleted file mode 100644 index c23d409f..00000000 --- a/gi/pipeline/clsp.config +++ /dev/null @@ -1,10 +0,0 @@ -# THIS FILE GIVES THE LOCATIONS OF THE CORPORA USED -# name path aligned-corpus LM dev dev-refs test1 testt-eval.sh ... -/export/ws10smt/data -btec /export/ws10smt/data/btec/ split.zh-en.al lm/en.3gram.lm.gz devtest/devset1_2.zh devtest/devset1_2.lc.en* devtest/devset3.zh eval-devset3.sh -fbis /export/ws10smt/data/chinese-english.fbis corpus.zh-en.al -zhen /export/ws10smt/data/chinese-english corpus.zh-en.al lm/c2e.3gram.lm.gz dev_and_test/mt02.src.txt dev_and_test/mt02.ref.* dev_and_test/mt03.src.txt eval-mt03.sh -aren /export/ws10smt/data/arabic-english corpus.ar-en-al lm/a2e.3gram.lm.gz dev_and_test/dev.src.txt dev_and_test/dev.ref.txt.* dev_and_test/mt05.src.txt eval-mt05.sh -uren /export/ws10smt/data/urdu-english corpus.ur-en.al lm/u2e.en.lm.gz dev/dev.ur dev/dev.en* devtest/devtest.ur eval-devtest.sh -nlfr /export/ws10smt/data/dutch-french corpus.nl-fr.al - diff --git a/gi/pipeline/evaluation-pipeline.pl b/gi/pipeline/evaluation-pipeline.pl deleted file mode 100755 index 4b4529d9..00000000 --- a/gi/pipeline/evaluation-pipeline.pl +++ /dev/null @@ -1,364 +0,0 @@ -#!/usr/bin/perl -w -use strict; -use Getopt::Long; -use Cwd; -my $CWD = getcwd; - -my $SCRIPT_DIR; BEGIN { use Cwd qw/ abs_path /; use File::Basename; $SCRIPT_DIR = dirname(abs_path($0)); push @INC, $SCRIPT_DIR, "$SCRIPT_DIR/../../environment"; } -use LocalConfig; - -my $JOBS = 15; -my $PMEM = "9G"; -my $NUM_TRANSLATIONS = 50; -my $GOAL = "S"; - -# featurize_grammar may add multiple features from a single feature extractor -# the key in this map is the extractor name, the value is a list of the extracted features -my $feat_map = { - "LogRuleCount" => [ "LogRuleCount", "SingletonRule" ] , -# "XFeatures" => [ "XFE","XEF" ] , - "XFeatures" => [ "XFE","XEF","LabelledEF","LabelledFE"], # ,"XE_Singleton","XF_Singleton"] , - "LabelledRuleConditionals" => [ "LabelledFE","LabelledEF" ] , - "LexProb" => [ "LexE2F", "LexF2E" ] , - "BackoffRule" => [ "BackoffRule" ] , - "RulePenalty" => [ "RulePenalty" ] , - "LHSProb" => [ "LHSProb" ] , - "LabellingShape" => [ "LabellingShape" ] , - "GenerativeProb" => [ "GenerativeProb" ] , -}; - -my %init_weights = qw( - EGivenF -0.735245 - FGivenE -0.219391 - Glue -0.306709 - GlueTop 0.0473331 - LanguageModel 2.40403 - LexE2F -0.266989 - LexF2E -0.550373 - LogECount -0.129853 - LogFCount -0.194037 - LogRuleCount 0.256706 - BackoffRule 0.5 - XFE -0.256706 - XEF -0.256706 - XF_Singleton -0.05 - XE_Singleton -0.8 - LabelledFE -0.256706 - LabelledEF -0.256706 - PassThrough -0.9304905 - SingletonE -3.04161 - SingletonF 0.0714027 - SingletonRule -0.889377 - WordPenalty -1.99495 - RulePenalty -0.1 - LabellingShape -0.1 - LHSProb -0.1 - GenerativeProb -0.1 -); - - -# these features are included by default -my @DEFAULT_FEATS = qw( PassThrough Glue GlueTop LanguageModel WordPenalty ); - - - -my $FILTERBYF = "$SCRIPT_DIR/scripts/filter-by-f.pl"; -my $CDEC = "$SCRIPT_DIR/../../decoder/cdec"; -my $PARALLELIZE = "$SCRIPT_DIR/../../vest/parallelize.pl"; -my $EXTOOLS = "$SCRIPT_DIR/../../extools"; -die "Can't find extools: $EXTOOLS" unless -e $EXTOOLS && -d $EXTOOLS; -my $VEST = "$SCRIPT_DIR/../../vest"; -die "Can't find vest: $VEST" unless -e $VEST && -d $VEST; -my $DISTVEST = "$VEST/dist-vest.pl"; -my $FILTER = "$EXTOOLS/filter_grammar"; -my $FEATURIZE = "$EXTOOLS/featurize_grammar"; -assert_exec($CDEC, $PARALLELIZE, $FILTER, $FEATURIZE, $DISTVEST, $FILTERBYF); - -my $numtopics = 25; - -my $config = "$SCRIPT_DIR/" . (lc environment_name()) . '.config'; -print STDERR "CORPORA CONFIGURATION: $config\n"; -open CONF, "<$config" or die "Can't read $config: $!"; -my %paths; -my %corpora; -my %lms; -my %devs; -my %devrefs; -my %tests; -my %testevals; -my $datadir; -print STDERR " LANGUAGE PAIRS:"; -while() { - chomp; - next if /^#/; - next if /^\s*$/; - s/^\s+//; - s/\s+$//; - if (! defined $datadir) { $datadir = $_; next; } - my ($name, $path, $corpus, $lm, $dev, $devref, @xtests) = split /\s+/; - $paths{$name} = $path; - $corpora{$name} = $corpus; - $lms{$name} = $lm; - $devs{$name} = $dev; - $devrefs{$name} = $devref; - $tests{$name} = $xtests[0]; - $testevals{$name} = $xtests[1]; - print STDERR " $name"; -} -print STDERR "\n"; - -my %langpairs = map { $_ => 1 } qw( btec zhen fbis aren uren nlfr ); - -my $outdir = "$CWD/exp"; -my $help; -my $FEATURIZER_OPTS = ''; -my $dataDir = '/export/ws10smt/data'; -my @features; -my $bkoffgram; -my $gluegram; -my $oovgram; -my $usefork; -my $lmorder = 3; -my $density; -if (GetOptions( - "backoff-grammar=s" => \$bkoffgram, - "density-prune=f" => \$density, - "glue-grammar=s" => \$gluegram, - "oov-grammar=s" => \$oovgram, - "data=s" => \$dataDir, - "pmem=s" => \$PMEM, - "n=i" => \$NUM_TRANSLATIONS, - "features=s@" => \@features, - "use-fork" => \$usefork, - "jobs=i" => \$JOBS, - "out-dir=s" => \$outdir, - "lmorder=i" => \$lmorder, - "goal=s" => \$GOAL, -) == 0 || @ARGV!=2 || $help) { - print_help(); - exit; -} -my $DENSITY_PRUNE = ''; -if ($density) { - $DENSITY_PRUNE = "--density-prune $density"; -} -if ($usefork) { $usefork="--use-fork"; } else { $usefork = ''; } -my @fkeys = keys %$feat_map; -die "You must specify one or more features with -f. Known features: @fkeys\n" unless scalar @features > 0; -my @xfeats; -for my $feat (@features) { - my $rs = $feat_map->{$feat}; - if (!defined $rs) { die "DON'T KNOW ABOUT FEATURE $feat\n"; } - my @xfs = @$rs; - @xfeats = (@xfeats, @xfs); - $FEATURIZER_OPTS .= " -f $feat" unless $feat eq "BackoffRule"; -} -print STDERR "X-FEATS: @xfeats\n"; - -my $lp = $ARGV[0]; -my $grammar = $ARGV[1]; -print STDERR " CORPUS REPO: $dataDir\n"; -print STDERR " LANGUAGE PAIR: $lp\n"; -die "I don't know about that language pair\n" unless $paths{$lp}; -my $corpdir = "$dataDir"; -if ($paths{$lp} =~ /^\//) { $corpdir = $paths{$lp}; } else { $corpdir .= '/' . $paths{$lp}; } -die "I can't find the corpora directory: $corpdir" unless -d $corpdir; -print STDERR " GRAMMAR: $grammar\n"; -my $LANG_MODEL = mydircat($corpdir, $lms{$lp}); -print STDERR " LM: $LANG_MODEL\n"; -my $CORPUS = mydircat($corpdir, $corpora{$lp}); -die "Can't find corpus: $CORPUS" unless -f $CORPUS; - -my $dev = mydircat($corpdir, $devs{$lp}); -my $drefs = $devrefs{$lp}; -die "Can't find dev: $dev\n" unless -f $dev; -die "Dev refs not set" unless $drefs; -$drefs = mydircat($corpdir, $drefs); - -my $test = mydircat($corpdir, $tests{$lp}); -my $teval = mydircat($corpdir, $testevals{$lp}); -#die "Can't find test: $test\n" unless -f $test; -#assert_exec($teval); - -`mkdir -p $outdir`; - -# CREATE INIT WEIGHTS -print STDERR "\nCREATING INITIAL WEIGHTS FILE: weights.init\n"; -my $weights = mydircat($outdir, "weights.init"); -write_random_weights_file($weights, @xfeats); - -my $bkoff_grmr; -my $glue_grmr; -if($bkoffgram) { - print STDERR "Placing backoff grammar…\n"; - $bkoff_grmr = mydircat($outdir, "backoff.scfg.gz"); - print STDERR "cp $bkoffgram $bkoff_grmr\n"; - safesystem(undef,"cp $bkoffgram $bkoff_grmr"); -} -if($gluegram) { - print STDERR "Placing glue grammar…\n"; - $glue_grmr = mydircat($outdir, "glue.bo.scfg.gz"); - print STDERR "cp $gluegram $glue_grmr\n"; - safesystem(undef,"cp $gluegram $glue_grmr"); -} - -# MAKE DEV -print STDERR "\nFILTERING FOR dev...\n"; -print STDERR "DEV: $dev (REFS=$drefs)\n"; -my $devgrammar = filter($grammar, $dev, 'dev', $outdir); -my $devini = mydircat($outdir, "cdec-dev.ini"); -write_cdec_ini($devini, $devgrammar); - - -# MAKE TEST -print STDERR "\nFILTERING FOR test...\n"; -print STDERR "TEST: $test (EVAL=$teval)\n"; -`mkdir -p $outdir`; -my $testgrammar = filter($grammar, $test, 'test', $outdir); -my $testini = mydircat($outdir, "cdec-test.ini"); -write_cdec_ini($testini, $testgrammar); - - -# VEST -print STDERR "\nMINIMUM ERROR TRAINING\n"; -my $tuned_weights = mydircat($outdir, 'weights.tuned'); -if (-f $tuned_weights) { - print STDERR "TUNED WEIGHTS $tuned_weights EXISTS: REUSING\n"; -} else { - my $cmd = "$DISTVEST $usefork $DENSITY_PRUNE --decode-nodes $JOBS --pmem=$PMEM --ref-files=$drefs --source-file=$dev --weights $weights $devini"; - print STDERR "MERT COMMAND: $cmd\n"; - `rm -rf $outdir/vest 2> /dev/null`; - chdir $outdir or die "Can't chdir to $outdir: $!"; - $weights = `$cmd`; - die "MERT reported non-zero exit code" unless $? == 0; - chomp $weights; - safesystem($tuned_weights, "cp $weights $tuned_weights"); - print STDERR "TUNED WEIGHTS: $tuned_weights\n"; - die "$tuned_weights is missing!" unless -f $tuned_weights; -} - -# DECODE -print STDERR "\nDECODE TEST SET\n"; -my $decolog = mydircat($outdir, "test-decode.log"); -my $testtrans = mydircat($outdir, "test.trans"); -my $cmd = "cat $test | $PARALLELIZE $usefork -j $JOBS -e $decolog -- $CDEC -c $testini -w $tuned_weights > $testtrans"; -safesystem($testtrans, $cmd) or die "Failed to decode test set!"; - - -# EVALUATE -print STDERR "\nEVALUATE TEST SET\n"; -print STDERR "TEST: $testtrans\n"; -$cmd = "$teval $testtrans"; -safesystem(undef, $cmd) or die "Failed to evaluate!"; -exit 0; - - -sub write_random_weights_file { - my ($file, @extras) = @_; - if (-f $file) { - print STDERR "$file exists - REUSING!\n"; - return; - } - open F, ">$file" or die "Can't write $file: $!"; - my @feats = (@DEFAULT_FEATS, @extras); - for my $feat (@feats) { - my $r = rand(0.4) + 0.8; - my $w = $init_weights{$feat} * $r; - if ($w == 0) { $w = 0.0001; print STDERR "WARNING: $feat had no initial weight!\n"; } - print F "$feat $w\n"; - } - close F; -} - -sub filter { - my ($grammar, $set, $name, $outdir) = @_; - my $out1 = mydircat($outdir, "$name.filt.gz"); - my $out2 = mydircat($outdir, "$name.f_feat.gz"); - my $outgrammar = mydircat($outdir, "$name.scfg.gz"); - if (-f $outgrammar) { print STDERR "$outgrammar exists - REUSING!\n"; } else { - my $cmd = "gunzip -c $grammar | $FILTER -t $set | gzip > $out1"; - safesystem($out1, $cmd) or die "Filtering failed."; - $cmd = "gunzip -c $out1 | $FEATURIZE $FEATURIZER_OPTS -g $out1 -c $CORPUS | gzip > $out2"; - safesystem($out2, $cmd) or die "Featurizing failed"; - $cmd = "$FILTERBYF $NUM_TRANSLATIONS $out2 $outgrammar"; - safesystem($outgrammar, $cmd) or die "Secondary filtering failed"; - } - return $outgrammar; -} - -sub mydircat { - my ($base, $suffix) = @_; - if ($suffix =~ /^\//) { return $suffix; } - my $res = $base . '/' . $suffix; - $res =~ s/\/\//\//g; - return $res; -} - -sub write_cdec_ini { - my ($filename, $grammar_path) = (@_); - open CDECINI, ">$filename" or die "Can't write $filename: $!"; - my $glue = ($gluegram ? "$glue_grmr" : "$datadir/glue/glue.scfg.gz"); - my $oov = ($oovgram ? "$oovgram" : "$datadir/oov.scfg.gz"); - print CDECINI <> 8; - if ($exitcode) { - print STDERR "Exit code: $exitcode\n"; - if (defined $output && -e $output) { printf STDERR "Removing $output\n"; `rm -rf $output`; } - } - return ! $exitcode; - } -} - -sub assert_exec { - my @files = @_; - for my $file (@files) { - die "Can't find $file - did you run make?\n" unless -e $file; - die "Can't execute $file" unless -e $file; - } -}; - diff --git a/gi/pipeline/local-gi-pipeline.pl b/gi/pipeline/local-gi-pipeline.pl deleted file mode 100755 index e31167a2..00000000 --- a/gi/pipeline/local-gi-pipeline.pl +++ /dev/null @@ -1,465 +0,0 @@ -#!/usr/bin/perl -w -use strict; -use File::Copy; - -my $SCRIPT_DIR; BEGIN { use Cwd qw/ abs_path cwd /; use File::Basename; $SCRIPT_DIR = dirname(abs_path($0)); push @INC, $SCRIPT_DIR; } - -use Getopt::Long "GetOptions"; - -my $GZIP = 'gzip'; -my $ZCAT = 'gunzip -c'; -my $SED = 'sed -e'; -my $BASE_PHRASE_MAX_SIZE = 10; -my $COMPLETE_CACHE = 1; -my $ITEMS_IN_MEMORY = 10000000; # cache size in extractors -my $NUM_TOPICS = 50; -my $NUM_TOPICS_COARSE; -my $NUM_TOPICS_FINE = $NUM_TOPICS; -my $NUM_SAMPLES = 1000; -my $CONTEXT_SIZE = 1; -my $BIDIR = 0; -my $TOPICS_CONFIG = "pyp-topics.conf"; -my $LANGUAGE = "target"; -my $LABEL_THRESHOLD = "0"; -my $PRESERVE_PHRASES; - -my $MODEL = "pyp"; -my $NUM_ITERS = 100; -my $PR_SCALE_P = 0; -my $PR_SCALE_C = 0; -my $PR_FLAGS = ""; -my $MORFMARK = ""; - -my $EXTOOLS = "$SCRIPT_DIR/../../extools"; -die "Can't find extools: $EXTOOLS" unless -e $EXTOOLS && -d $EXTOOLS; -my $PYPTOOLS = "$SCRIPT_DIR/../pyp-topics/src"; -die "Can't find pyp-topics: $PYPTOOLS" unless -e $PYPTOOLS && -d $PYPTOOLS; -my $PYPSCRIPTS = "$SCRIPT_DIR/../pyp-topics/scripts"; -die "Can't find pyp-topics: $PYPSCRIPTS" unless -e $PYPSCRIPTS && -d $PYPSCRIPTS; -my $PRTOOLS = "$SCRIPT_DIR/../posterior-regularisation"; -die "Can't find posterior-regularisation: $PRTOOLS" unless -e $PRTOOLS && -d $PRTOOLS; -my $REDUCER = "$EXTOOLS/mr_stripe_rule_reduce"; -my $C2D = "$PYPSCRIPTS/contexts2documents.py"; -my $S2L = "$PYPSCRIPTS/spans2labels.py"; -my $SPLIT = "$SCRIPT_DIR/../posterior-regularisation/split-languages.py"; - -my $PREM_TRAIN="$PRTOOLS/prjava/train-PR-cluster.sh"; - -my $SORT_KEYS = "$SCRIPT_DIR/scripts/sort-by-key.sh"; -my $PATCH_CORPUS = "$SCRIPT_DIR/scripts/patch-corpus.pl"; -my $REMOVE_TAGS_CORPUS = "$SCRIPT_DIR/scripts/remove-tags-from-corpus.pl"; -my $REMOVE_TAGS_CONTEXT = "$SCRIPT_DIR/scripts/remove-tags-from-contexts.pl"; -my $EXTRACTOR = "$EXTOOLS/extractor"; -my $TOPIC_TRAIN = "$PYPTOOLS/pyp-contexts-train"; -my $MORF_DOC_FILTER = "$SCRIPT_DIR/../morf-segmentation/filter_docs.pl"; - -assert_exec($PATCH_CORPUS, $SORT_KEYS, $REDUCER, $EXTRACTOR, - $S2L, $C2D, $TOPIC_TRAIN, $SPLIT, $REMOVE_TAGS_CONTEXT, $REMOVE_TAGS_CORPUS, $MORF_DOC_FILTER); - -my $BACKOFF_GRAMMAR; -my $DEFAULT_CAT; -my $HIER_CAT; -my %FREQ_HIER = (); -my $TAGGED_CORPUS; - -my $NAME_SHORTCUT; - -my $OUTPUT = './giwork'; -usage() unless &GetOptions('base_phrase_max_size=i' => \$BASE_PHRASE_MAX_SIZE, - 'backoff_grammar' => \$BACKOFF_GRAMMAR, - 'output=s' => \$OUTPUT, - 'model=s' => \$MODEL, - 'topics=i' => \$NUM_TOPICS_FINE, - 'coarse_topics=i' => \$NUM_TOPICS_COARSE, - 'trg_context=i' => \$CONTEXT_SIZE, - 'samples=i' => \$NUM_SAMPLES, - 'label_threshold=f' => \$LABEL_THRESHOLD, - 'use_default_cat' => \$DEFAULT_CAT, - 'topics-config=s' => \$TOPICS_CONFIG, - 'iterations=i' => \$NUM_ITERS, - 'pr-scale-phrase=f' => \$PR_SCALE_P, - 'pr-scale-context=f' => \$PR_SCALE_C, - 'pr-flags=s' => \$PR_FLAGS, - 'tagged_corpus=s' => \$TAGGED_CORPUS, - 'language=s' => \$LANGUAGE, - 'get_name_only' => \$NAME_SHORTCUT, - 'preserve_phrases' => \$PRESERVE_PHRASES, - 'morf=s' => \$MORFMARK, - ); -if ($NAME_SHORTCUT) { - $NUM_TOPICS = $NUM_TOPICS_FINE; - print STDERR labeled_dir(); - exit 0; -} -usage() unless scalar @ARGV == 1; -my $CORPUS = $ARGV[0]; -open F, "<$CORPUS" or die "Can't read $CORPUS: $!"; close F; - -$NUM_TOPICS = $NUM_TOPICS_FINE; - -$HIER_CAT = ( $NUM_TOPICS_COARSE ? 1 : 0 ); - -print STDERR " Output: $OUTPUT\n"; -my $DATA_DIR = $OUTPUT . '/corpora'; -my $LEX_NAME = "corpus.f_e_a.$LANGUAGE.lex"; -my $CORPUS_LEX = $DATA_DIR . '/' . $LEX_NAME; # corpus used to extract rules -my $CORPUS_CLUSTER = $DATA_DIR . "/corpus.f_e_a.$LANGUAGE.cluster"; # corpus used for clustering (often identical) - -my $CONTEXT_DIR = $OUTPUT . '/' . context_dir(); -my $CLUSTER_DIR = $OUTPUT . '/' . cluster_dir(); -my $LABELED_DIR = $OUTPUT . '/' . labeled_dir(); -my $CLUSTER_DIR_C; -my $CLUSTER_DIR_F; -my $LABELED_DIR_C; -my $LABELED_DIR_F; -if($HIER_CAT) { - $CLUSTER_DIR_F = $CLUSTER_DIR; - $LABELED_DIR_F = $LABELED_DIR; - $NUM_TOPICS = $NUM_TOPICS_COARSE; - $CLUSTER_DIR_C = $OUTPUT . '/' . cluster_dir(); - $LABELED_DIR_C = $OUTPUT . '/' . labeled_dir(); - $NUM_TOPICS = $NUM_TOPICS_FINE; -} -my $GRAMMAR_DIR = $OUTPUT . '/' . grammar_dir(); -print STDERR " Context: $CONTEXT_DIR\n Cluster: $CLUSTER_DIR\n Labeled: $LABELED_DIR\n Grammar: $GRAMMAR_DIR\n"; -safemkdir($OUTPUT) or die "Couldn't create output directory $OUTPUT: $!"; -safemkdir($DATA_DIR) or die "Couldn't create output directory $DATA_DIR: $!"; -safemkdir($CONTEXT_DIR) or die "Couldn't create output directory $CONTEXT_DIR: $!"; -safemkdir($CLUSTER_DIR) or die "Couldn't create output directory $CLUSTER_DIR: $!"; -if($HIER_CAT) { - safemkdir($CLUSTER_DIR_C) or die "Couldn't create output directory $CLUSTER_DIR_C: $!"; - safemkdir($LABELED_DIR_C) or die "Couldn't create output directory $LABELED_DIR_C: $!"; -} -safemkdir($LABELED_DIR) or die "Couldn't create output directory $LABELED_DIR: $!"; -safemkdir($GRAMMAR_DIR) or die "Couldn't create output directory $GRAMMAR_DIR: $!"; -if(-e $TOPICS_CONFIG) { - copy($TOPICS_CONFIG, $CLUSTER_DIR) or die "Copy failed: $!"; -} - -setup_data(); - -if (lc($MODEL) eq "blagree") { - extract_bilingual_context(); -} else { - extract_context(); -} - -if (lc($MODEL) eq "pyp") { - if($HIER_CAT) { - $NUM_TOPICS = $NUM_TOPICS_COARSE; - $CLUSTER_DIR = $CLUSTER_DIR_C; - topic_train(); - $NUM_TOPICS = $NUM_TOPICS_FINE; - $CLUSTER_DIR = $CLUSTER_DIR_F; - topic_train(); - } else { - topic_train(); - } -} elsif (lc($MODEL) =~ /pr|em|agree/) { - prem_train(); -} else { die "Unsupported model type: $MODEL. Must be one of PYP or PREM.\n"; } -if($HIER_CAT) { - $NUM_TOPICS = $NUM_TOPICS_COARSE; - $CLUSTER_DIR = $CLUSTER_DIR_C; - $LABELED_DIR = $LABELED_DIR_C; - label_spans_with_topics(); - $NUM_TOPICS = $NUM_TOPICS_FINE; - $CLUSTER_DIR = $CLUSTER_DIR_F; - $LABELED_DIR = $LABELED_DIR_F; - label_spans_with_topics(); - extract_freqs(); -} else { - label_spans_with_topics(); -} -my $res; -if ($BIDIR) { - $res = grammar_extract_bidir(); -} else { - $res = grammar_extract(); -} -print STDERR "\n!!!COMPLETE!!!\n"; -print STDERR "GRAMMAR: $res\nYou should probably run: $SCRIPT_DIR/evaluation-pipeline.pl LANGPAIR giwork/ct1s0.L10.PYP.t4.s20.grammar/grammar.gz -f FEAT1 -f FEAT2\n\n"; -exit 0; - -sub setup_data { - print STDERR "\n!!!PREPARE CORPORA!!!\n"; - if (-f $CORPUS_LEX && $CORPUS_CLUSTER) { - print STDERR "$CORPUS_LEX and $CORPUS_CLUSTER exist, reusing...\n"; - return; - } - copy($CORPUS, $CORPUS_LEX); - if ($TAGGED_CORPUS) { - die "Can't find $TAGGED_CORPUS" unless -f $TAGGED_CORPUS; - my $opt=""; - $opt = "-s" if ($LANGUAGE eq "source"); - $opt = $opt . " -a" if ($PRESERVE_PHRASES); - my $cmd="$PATCH_CORPUS $opt $TAGGED_CORPUS $CORPUS_LEX > $CORPUS_CLUSTER"; - safesystem($cmd) or die "Failed to extract contexts."; - } else { - symlink($LEX_NAME, $CORPUS_CLUSTER); - } -} - -sub context_dir { - return "ct${CONTEXT_SIZE}s0.L$BASE_PHRASE_MAX_SIZE.l$LANGUAGE"; -} - -sub cluster_dir { - if (lc($MODEL) eq "pyp") { - return context_dir() . ".PYP.t$NUM_TOPICS.s$NUM_SAMPLES"; - } elsif (lc($MODEL) eq "em") { - return context_dir() . ".EM.t$NUM_TOPICS.i$NUM_ITERS"; - } elsif (lc($MODEL) eq "pr") { - return context_dir() . ".PR.t$NUM_TOPICS.i$NUM_ITERS.sp$PR_SCALE_P.sc$PR_SCALE_C"; - } elsif (lc($MODEL) eq "agree") { - return context_dir() . ".AGREE.t$NUM_TOPICS.i$NUM_ITERS"; - } elsif (lc($MODEL) eq "blagree") { - return context_dir() . ".BLAGREE.t$NUM_TOPICS.i$NUM_ITERS"; - } -} - -sub labeled_dir { - if (lc($MODEL) eq "pyp" && $LABEL_THRESHOLD ne "0") { - return cluster_dir() . "_lt$LABEL_THRESHOLD"; - } else { - return cluster_dir(); - } -} - -sub grammar_dir { - # TODO add grammar config options -- adjacent NTs, etc - if($HIER_CAT) { - return cluster_dir() . ".hier$NUM_TOPICS_COARSE-$NUM_TOPICS_FINE.grammar"; - } else { - return labeled_dir() . ".grammar"; - } -} - - - -sub safemkdir { - my $dir = shift; - if (-d $dir) { return 1; } - return mkdir($dir); -} - -sub usage { - print < $CLUSTER_DIR/clusters.txt") or die "Failed to unzip"; - safesystem("$EXTRACTOR --base_phrase_spans -i $CORPUS_CLUSTER -c $ITEMS_IN_MEMORY -L $BASE_PHRASE_MAX_SIZE -S $CONTEXT_SIZE | $S2L $CLUSTER_DIR/clusters.txt $CONTEXT_SIZE $LABEL_THRESHOLD $extra > $OUT_SPANS") or die "Failed to label spans"; - unlink("$CLUSTER_DIR/clusters.txt") or warn "Failed to remove $CLUSTER_DIR/clusters.txt"; - safesystem("paste -d ' ' $CORPUS_LEX $OUT_SPANS | sed 's/ *||| *\$//' > $LABELED_DIR/corpus.src_trg_al_label") or die "Couldn't paste"; - } -} - -sub extract_freqs { - print STDERR "\n!!!EXTRACTING FREQUENCIES\n"; - my $IN_COARSE = "$LABELED_DIR_C/labeled_spans.txt"; - my $IN_FINE = "$LABELED_DIR_F/labeled_spans.txt"; - my $OUT_SPANS = "$LABELED_DIR_F/labeled_spans.hier$NUM_TOPICS_COARSE-$NUM_TOPICS_FINE.txt"; - my $FREQS = "$LABELED_DIR_F/label_freq.hier$NUM_TOPICS_COARSE-$NUM_TOPICS_FINE.txt"; - my $COARSE_EXPR = "\'s/\\(X[0-9][0-9]*\\)/\\1c/g\'"; #' - my $FINE_EXPR = "\'s/\\(X[0-9][0-9]*\\)/\\1f/g\'"; #' - my %finehier = (); - if (-e $OUT_SPANS) { - print STDERR "$OUT_SPANS exists, reusing...\n"; - } else { - safesystem("paste -d ' ' $IN_COARSE $IN_FINE > $OUT_SPANS"); - } - open SPANS, $OUT_SPANS or die $!; - while () { - my ($tmp, $coarse, $fine) = split /\|\|\|/; - my @coarse_spans = $coarse =~ /\d+-\d+:X(\d+)/g; - my @fine_spans = $fine =~ /\d+-\d+:X(\d+)/g; - - foreach my $i (0..(scalar @coarse_spans)-1) { - my $coarse_cat = $coarse_spans[$i]; - my $fine_cat = $fine_spans[$i]; - - $FREQ_HIER{$coarse_cat}{$fine_cat}++; - } - } - close SPANS; - foreach (values %FREQ_HIER) { - my $coarse_freq = $_; - my $total = 0; - $total+=$_ for (values %{ $coarse_freq }); - $coarse_freq->{$_}=log($coarse_freq->{$_}/$total) for (keys %{ $coarse_freq }); - } - open FREQS, ">", $FREQS or die $!; - foreach my $coarse_cat (keys %FREQ_HIER) { - print FREQS "$coarse_cat |||"; - foreach my $fine_cat (keys %{$FREQ_HIER{$coarse_cat}}) { - my $res = $FREQ_HIER{$coarse_cat}{$fine_cat}; - print FREQS " $fine_cat:$res"; - if(! exists $finehier{$fine_cat} || $finehier{$fine_cat} < $res) { - $finehier{$fine_cat} = $coarse_cat; - } - } - print FREQS "\n"; - } -# foreach my $fine_cat (keys %finehier) { -# print FREQS "$fine_cat -> $finehier{$fine_cat}\n"; -# } - close FREQS; - $CLUSTER_DIR = $CLUSTER_DIR_F; -} - -sub grammar_extract { - my $LABELED = "$LABELED_DIR/corpus.src_trg_al_label"; - print STDERR "\n!!!EXTRACTING GRAMMAR\n"; - my $OUTGRAMMAR = "$GRAMMAR_DIR/grammar.gz"; - if (-e $OUTGRAMMAR) { - print STDERR "$OUTGRAMMAR exists, reusing...\n"; - } else { - my $BACKOFF_ARG = ($BACKOFF_GRAMMAR ? "-g" : ""); - my $DEFAULT_CAT_ARG = ($DEFAULT_CAT ? "-d X" : ""); - safesystem("$EXTRACTOR -i $LABELED -c $ITEMS_IN_MEMORY -L $BASE_PHRASE_MAX_SIZE -t $NUM_TOPICS $BACKOFF_ARG $DEFAULT_CAT_ARG | $SORT_KEYS | $REDUCER -p | $GZIP > $OUTGRAMMAR") or die "Couldn't extract grammar"; - } - return $OUTGRAMMAR; -} - -sub grammar_extract_bidir { -#gzcat ex.output.gz | ./mr_stripe_rule_reduce -p -b | sort -t $'\t' -k 1 | ./mr_stripe_rule_reduce | gzip > phrase-table.gz - my $LABELED = "$LABELED_DIR/corpus.src_trg_al_label"; - print STDERR "\n!!!EXTRACTING GRAMMAR\n"; - my $OUTGRAMMAR = "$GRAMMAR_DIR/grammar.bidir.gz"; - if (-e $OUTGRAMMAR) { - print STDERR "$OUTGRAMMAR exists, reusing...\n"; - } else { - my $BACKOFF_ARG = ($BACKOFF_GRAMMAR ? "-g" : ""); - safesystem("$EXTRACTOR -i $LABELED -c $ITEMS_IN_MEMORY -L $BASE_PHRASE_MAX_SIZE -b -t $NUM_TOPICS $BACKOFF_ARG | $SORT_KEYS | $REDUCER -p -b | $SORT_KEYS | $REDUCER | $GZIP > $OUTGRAMMAR") or die "Couldn't extract grammar"; - } - return $OUTGRAMMAR; -} - -sub safesystem { - print STDERR "Executing: @_\n"; - system(@_); - if ($? == -1) { - print STDERR "ERROR: Failed to execute: @_\n $!\n"; - exit(1); - } - elsif ($? & 127) { - printf STDERR "ERROR: Execution of: @_\n died with signal %d, %s coredump\n", - ($? & 127), ($? & 128) ? 'with' : 'without'; - exit(1); - } - else { - my $exitcode = $? >> 8; - print STDERR "Exit code: $exitcode\n" if $exitcode; - return ! $exitcode; - } -} - diff --git a/gi/pipeline/lticluster.config b/gi/pipeline/lticluster.config deleted file mode 100644 index 3e23c8cb..00000000 --- a/gi/pipeline/lticluster.config +++ /dev/null @@ -1,9 +0,0 @@ -# THIS FILE GIVES THE LOCATIONS OF THE CORPORA USED -# name path aligned-corpus LM dev dev-refs test1 testt-eval.sh ... -/home/cdyer/ws10smt-data -btec /home/cdyer/ws10smt-data/btec/ split.zh-en.al lm/en.3gram.lm.gz devtest/devset1_2.zh devtest/devset1_2.lc.en* devtest/devset3.zh eval-devset3.sh -zhen /home/cdyer/ws10smt-data/chinese-english corpus.zh-en.al lm/c2e.3gram.lm.gz dev_and_test/mt02.src.txt dev_and_test/mt02.ref.* dev_and_test/mt03.src.txt eval-mt03.sh -aren /home/cdyer/ws10smt-data/arabic-english corpus.ar-en-al lm/a2e.3gram.lm.gz dev_and_test/dev.src.txt dev_and_test/dev.ref.txt.* dev_and_test/mt05.src.txt eval-mt05.sh -uren /home/cdyer/ws10smt-data/urdu-english corpus.ur-en.al lm/u2e.en.lm.gz dev/dev.ur dev/dev.en* devtest/devtest.ur eval-devtest.sh -nlfr /home/cdyer/ws10smt-data/dutch-french corpus.nl-fr.al - diff --git a/gi/pipeline/scripts/filter-by-f.pl b/gi/pipeline/scripts/filter-by-f.pl deleted file mode 100755 index 0cef0606..00000000 --- a/gi/pipeline/scripts/filter-by-f.pl +++ /dev/null @@ -1,56 +0,0 @@ -#!/usr/bin/perl -w -use strict; - -my $SCRIPT_DIR; BEGIN { use Cwd qw/ abs_path /; use File::Basename; $SCRIPT_DIR = dirname(abs_path($0)); push @INC, $SCRIPT_DIR; } - -my $REKEY="$SCRIPT_DIR/rekey.pl"; -my $REFILTER="$SCRIPT_DIR/refilter.pl"; -my $SORT="$SCRIPT_DIR/sort-by-key.sh"; -assert_exec($REKEY, $REFILTER, $SORT); - - -die "Usage: $0 NUM-TRANSLATIONS ingrammar.gz outgrammar.gz\n" unless scalar @ARGV == 3; -my $translations = shift @ARGV; -die "Need number: $translations" unless $translations > 0; -die unless $ARGV[0] =~ /\.gz$/; -die unless $ARGV[1] =~ /\.gz$/; -die if $ARGV[0] eq $ARGV[1]; -die "Can't find $ARGV[0]" unless -f $ARGV[0]; - -my $cmd = "gunzip -c $ARGV[0] | $REKEY | $SORT | $REFILTER $translations | gzip > $ARGV[1]"; -safesystem($ARGV[1], $cmd) or die "Filtering failed"; -exit 0; - -sub assert_exec { - my @files = @_; - for my $file (@files) { - die "Can't find $file - did you run make?\n" unless -e $file; - die "Can't execute $file" unless -e $file; - } -}; - -sub safesystem { - my $output = shift @_; - print STDERR "Executing: @_\n"; - system(@_); - if ($? == -1) { - print STDERR "ERROR: Failed to execute: @_\n $!\n"; - if (defined $output && -e $output) { printf STDERR "Removing $output\n"; `rm -rf $output`; } - exit(1); - } - elsif ($? & 127) { - printf STDERR "ERROR: Execution of: @_\n died with signal %d, %s coredump\n", - ($? & 127), ($? & 128) ? 'with' : 'without'; - if (defined $output && -e $output) { printf STDERR "Removing $output\n"; `rm -rf $output`; } - exit(1); - } - else { - my $exitcode = $? >> 8; - if ($exitcode) { - print STDERR "Exit code: $exitcode\n"; - if (defined $output && -e $output) { printf STDERR "Removing $output\n"; `rm -rf $output`; } - } - return ! $exitcode; - } -} - diff --git a/gi/pipeline/scripts/patch-corpus.pl b/gi/pipeline/scripts/patch-corpus.pl deleted file mode 100755 index c0eec43e..00000000 --- a/gi/pipeline/scripts/patch-corpus.pl +++ /dev/null @@ -1,65 +0,0 @@ -#!/usr/bin/perl -w -use strict; - -my $PATCH = shift @ARGV; -my $TGT = 1; -my $APPEND; -while ($PATCH eq "-s" || $PATCH eq "-a") { - if ($PATCH eq "-s") { - undef $TGT; - } else { - $APPEND = 1; - } - $PATCH = shift @ARGV; -} - -die "Usage: $0 [-s] [-a] tagged.en[_fr] < lexical.en_fr_al[_...]\n" unless $PATCH; - -open P, "<$PATCH" or die "Can't read tagged corpus $PATCH: $!"; -my $first=

; close P; -my @fields = split / \|\|\| /, $first; -die "Bad format!" if (scalar @fields > 2); - -if (scalar @fields != 1) { - # TODO support this - die "Patching source and target not supported yet!"; -} - -my $line = 0; -open P, "<$PATCH" or die "Can't read tagged corpus $PATCH: $!"; -while(my $pline =

) { - chomp $pline; - $line++; - my $line = <>; - die "Too few lines in lexical corpus!" unless $line; - chomp $line; - @fields = split / \|\|\| /, $line; - my @pwords = split /\s+/, $pline; - if ($TGT) { - my @lwords = split /\s+/, $fields[1]; - die "Length mismatch in line $line!\n" unless (scalar @pwords == scalar @lwords); - if ($APPEND) { - foreach my $i (0..(scalar @pwords-1)) { - $lwords[$i] = $lwords[$i] . '_' . $pwords[$i]; - } - $fields[1] = join ' ', @lwords; - } else { - $fields[1] = $pline; - } - } else { # source side - my @lwords = split /\s+/, $fields[0]; - die "Length mismatch in line $line!\n" unless (scalar @pwords == scalar @lwords); - if ($APPEND) { - foreach my $i (0..(scalar @pwords-1)) { - $lwords[$i] = $lwords[$i] . '_' . $pwords[$i]; - } - $fields[0] = join ' ', @lwords; - } else { - $fields[0] = $pline; - } - } - print join ' ||| ', @fields; - print "\n"; -} - - diff --git a/gi/pipeline/scripts/refilter.pl b/gi/pipeline/scripts/refilter.pl deleted file mode 100755 index a783eb4e..00000000 --- a/gi/pipeline/scripts/refilter.pl +++ /dev/null @@ -1,40 +0,0 @@ -#!/usr/bin/perl -w -use strict; - -my $NUM_TRANSLATIONS = shift @ARGV; -unless ($NUM_TRANSLATIONS) { $NUM_TRANSLATIONS=30; } -print STDERR "KEEPING $NUM_TRANSLATIONS TRANSLATIONS FOR SOURCE\n"; - -my $pk = ''; -my %dict; -while(<>) { - s/^(.+)\t//; - my $key = $1; - if ($key ne $pk) { - if ($pk) { - emit_dict(); - } - %dict = (); - $pk = $key; - } - my ($lhs, $f, $e, $s) = split / \|\|\| /; - my $score = 0; - if ($s =~ /XEF=([^ ]+)/) { - $score += $1; - } else { die; } - if ($s =~ /GenerativeProb=([^ ]+)/) { - $score += ($1 / 10); - } else { die; } - $dict{"$lhs ||| $f ||| $e ||| $s"} = $score; -} -emit_dict(); - -sub emit_dict { - my $cc = 0; - for my $k (sort { $dict{$a} <=> $dict{$b} } keys %dict) { - print "$k"; - $cc++; - if ($cc >= $NUM_TRANSLATIONS) { last; } - } -} - diff --git a/gi/pipeline/scripts/rekey.pl b/gi/pipeline/scripts/rekey.pl deleted file mode 100755 index 31eb86b8..00000000 --- a/gi/pipeline/scripts/rekey.pl +++ /dev/null @@ -1,8 +0,0 @@ -#!/usr/bin/perl - -while(<>) { - my ($lhs, $f, $e, $s) = split / \|\|\| /; - $f =~ s/\[X[0-9]+\]/\[X\]/g; - print "$f\t$_"; -} - diff --git a/gi/pipeline/scripts/remove-tags-from-contexts.pl b/gi/pipeline/scripts/remove-tags-from-contexts.pl deleted file mode 100755 index 20698816..00000000 --- a/gi/pipeline/scripts/remove-tags-from-contexts.pl +++ /dev/null @@ -1,53 +0,0 @@ -#!/usr/bin/perl -w -use strict; - -use Getopt::Long "GetOptions"; - -my $PHRASE = 'tok'; -my $CONTEXT = 'tag'; - -die "Usage: $0 [--phrase=tok|tag] [--context=tok|tag] < corpus" - unless &GetOptions('phrase=s' => \$PHRASE, 'context=s' => \$CONTEXT); - -my $lno = 0; -while(my $line = <>) { - $lno++; - chomp $line; - my @top = split /\t/, $line; - die unless (scalar @top == 2); - - my @pwords = split /\s+/, $top[0]; - foreach my $token (@pwords) { - #print $token . "\n"; - my @parts = split /_(?!.*_)/, $token; - die unless (scalar @parts == 2); - if ($PHRASE eq "tok") { - $token = $parts[0] - } elsif ($PHRASE eq "tag") { - $token = $parts[1] - } - } - - my @fields = split / \|\|\| /, $top[1]; - foreach my $i (0..((scalar @fields) / 2 - 1)) { - #print $i . ": " . $fields[2*$i] . " of " . (scalar @fields) . "\n"; - my @cwords = split /\s+/, $fields[2*$i]; - foreach my $token (@cwords) { - #print $i . ": " . $token . "\n"; - my @parts = split /_(?!.*_)/, $token; - if (scalar @parts == 2) { - if ($CONTEXT eq "tok") { - $token = $parts[0] - } elsif ($CONTEXT eq "tag") { - $token = $parts[1] - } - } - } - $fields[2*$i] = join ' ', @cwords; - } - - print join ' ', @pwords; - print "\t"; - print join ' ||| ', @fields; - print "\n"; -} diff --git a/gi/pipeline/scripts/remove-tags-from-corpus.pl b/gi/pipeline/scripts/remove-tags-from-corpus.pl deleted file mode 100755 index be3e97c0..00000000 --- a/gi/pipeline/scripts/remove-tags-from-corpus.pl +++ /dev/null @@ -1,44 +0,0 @@ -#!/usr/bin/perl -w -use strict; - -use Getopt::Long "GetOptions"; - -my $LANGUAGE = shift @ARGV; -$LANGUAGE = 'target' unless ($LANGUAGE); - -my $lno = 0; -while(my $line = <>) { - $lno++; - chomp $line; - - my @fields = split / \|\|\| /, $line; - - if ($LANGUAGE eq "source" or $LANGUAGE eq "both") { - my @cwords = split /\s+/, $fields[0]; - foreach my $token (@cwords) { - my @parts = split /_(?!.*_)/, $token; - if (scalar @parts == 2) { - $token = $parts[0] - } else { - print STDERR "WARNING: invalid tagged token $token\n"; - } - } - $fields[0] = join ' ', @cwords; - } - - if ($LANGUAGE eq "target" or $LANGUAGE eq "both") { - my @cwords = split /\s+/, $fields[1]; - foreach my $token (@cwords) { - my @parts = split /_(?!.*_)/, $token; - if (scalar @parts == 2) { - $token = $parts[1] - } else { - print STDERR "WARNING: invalid tagged token $token\n"; - } - } - $fields[0] = join ' ', @cwords; - } - - print join ' ||| ', @fields; - print "\n"; -} diff --git a/gi/pipeline/scripts/sort-by-key.sh b/gi/pipeline/scripts/sort-by-key.sh deleted file mode 100755 index 7ae33e03..00000000 --- a/gi/pipeline/scripts/sort-by-key.sh +++ /dev/null @@ -1,5 +0,0 @@ -#!/bin/bash - -export LANG=C -sort -t $'\t' -k 1 -T /tmp -S 6000000000 - diff --git a/gi/pipeline/scripts/xfeats.pl b/gi/pipeline/scripts/xfeats.pl deleted file mode 100755 index dc578513..00000000 --- a/gi/pipeline/scripts/xfeats.pl +++ /dev/null @@ -1,39 +0,0 @@ -#!/usr/bin/perl -w -use strict; - -die "Usage: $0 x-grammar.scfg[.gz] < cat-grammar.scfg\n" unless scalar @ARGV > 0; - -my $xgrammar = shift @ARGV; -die "Can't find $xgrammar" unless -f $xgrammar; -my $fh; -if ($xgrammar =~ /\.gz$/) { - open $fh, "gunzip -c $xgrammar|" or die "Can't fork: $!"; -} else { - open $fh, "<$xgrammar" or die "Can't read $xgrammar: $!"; -} -print STDERR "Reading X-feats from $xgrammar...\n"; -my %dict; -while(<$fh>) { - chomp; - my ($lhs, $f, $e, $feats) = split / \|\|\| /; - my $xfeats; - my $cc = 0; - my @xfeats = (); - while ($feats =~ /(EGivenF|FGivenE|LogRuleCount|LogECount|LogFCount|SingletonRule|SingletonE|SingletonF)=([^ ]+)( |$)/og) { - push @xfeats, "X_$1=$2"; - } - #print "$lhs ||| $f ||| $e ||| @xfeats\n"; - $dict{"$lhs ||| $f ||| $e"} = "@xfeats"; -} -close $fh; - -print STDERR "Add features...\n"; -while(<>) { - chomp; - my ($lhs, $f, $e) = split / \|\|\| /; - $f=~ s/\[[^]]+,([12])\]/\[X,$1\]/g; - my $xfeats = $dict{"[X] ||| $f ||| $e"}; - die "Can't find x features for: $_\n" unless $xfeats; - print "$_ $xfeats\n"; -} - diff --git a/gi/pipeline/valhalla.config b/gi/pipeline/valhalla.config deleted file mode 100644 index e00a8485..00000000 --- a/gi/pipeline/valhalla.config +++ /dev/null @@ -1,9 +0,0 @@ -# THIS FILE GIVES THE LOCATIONS OF THE CORPORA USED -# name path aligned-corpus LM dev dev-refs test1 testt-eval.sh ... -/home/chris/ws10smt/data -btec /home/chris/ws10smt/data/btec/ split.zh-en.al lm/en.3gram.lm.gz devtest/devset1_2.zh devtest/devset1_2.lc.en* devtest/devset3.zh eval-devset3.sh -fbis /home/chris/ws10smt/data/chinese-english.fbis corpus.zh-en.al -zhen /home/chris/ws10smt/data/chinese-english corpus.zh-en.al -aren /home/chris/ws10smt/data/arabic-english corpus.ar-en.al -uren /home/chris/ws10smt/data/urdu-english corpus.ur-en.al lm/u2e.en.lm.gz dev/dev.ur dev/dev.en* devtest/devtest.ur eval-devtest.sh -nlfr /home/chris/ws10smt/data/dutch-french corpus.nl-fr.al -- cgit v1.2.3