From 81578ddd4a32ee06d964bd7b5740ca61f76d5bc1 Mon Sep 17 00:00:00 2001 From: Chris Dyer Date: Thu, 26 Apr 2012 00:06:09 -0400 Subject: mostly implemented rampion optimizer --- rampion/Makefile.am | 6 + rampion/rampion.pl | 526 ++++++++++++++++++++++++++++++++++++++ rampion/rampion_cccp.cc | 157 ++++++++++++ rampion/rampion_generate_input.pl | 18 ++ 4 files changed, 707 insertions(+) create mode 100644 rampion/Makefile.am create mode 100755 rampion/rampion.pl create mode 100644 rampion/rampion_cccp.cc create mode 100755 rampion/rampion_generate_input.pl (limited to 'rampion') diff --git a/rampion/Makefile.am b/rampion/Makefile.am new file mode 100644 index 00000000..12df39c2 --- /dev/null +++ b/rampion/Makefile.am @@ -0,0 +1,6 @@ +bin_PROGRAMS = rampion_cccp + +rampion_cccp_SOURCES = rampion_cccp.cc +rampion_cccp_LDADD = $(top_srcdir)/decoder/libcdec.a $(top_srcdir)/mteval/libmteval.a $(top_srcdir)/utils/libutils.a -lz + +AM_CPPFLAGS = -W -Wall -Wno-sign-compare $(GTEST_CPPFLAGS) -I$(top_srcdir)/utils -I$(top_srcdir)/decoder -I$(top_srcdir)/mteval -I$(top_srcdir)/training diff --git a/rampion/rampion.pl b/rampion/rampion.pl new file mode 100755 index 00000000..fda2bac2 --- /dev/null +++ b/rampion/rampion.pl @@ -0,0 +1,526 @@ +#!/usr/bin/env perl +use strict; +my @ORIG_ARGV=@ARGV; +use Cwd qw(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"; } + +# Skip local config (used for distributing jobs) if we're running in local-only mode +use LocalConfig; +use Getopt::Long; +use IPC::Open2; +use POSIX ":sys_wait_h"; +my $QSUB_CMD = qsub_args(mert_memory()); +my $default_jobs = env_default_jobs(); + +my $VEST_DIR="$SCRIPT_DIR/../dpmert"; +require "$VEST_DIR/libcall.pl"; + +# Default settings +my $srcFile; +my $refFiles; +my $bin_dir = $SCRIPT_DIR; +die "Bin directory $bin_dir missing/inaccessible" unless -d $bin_dir; +my $FAST_SCORE="$bin_dir/../mteval/fast_score"; +die "Can't execute $FAST_SCORE" unless -x $FAST_SCORE; +my $MAPINPUT = "$bin_dir/rampion_generate_input.pl"; +my $MAPPER = "$bin_dir/rampion_cccp"; +my $parallelize = "$VEST_DIR/parallelize.pl"; +my $libcall = "$VEST_DIR/libcall.pl"; +my $sentserver = "$VEST_DIR/sentserver"; +my $sentclient = "$VEST_DIR/sentclient"; +my $LocalConfig = "$SCRIPT_DIR/../environment/LocalConfig.pm"; + +my $SCORER = $FAST_SCORE; +die "Can't find $MAPPER" unless -x $MAPPER; +my $cdec = "$bin_dir/../decoder/cdec"; +die "Can't find decoder in $cdec" unless -x $cdec; +die "Can't find $parallelize" unless -x $parallelize; +die "Can't find $libcall" unless -e $libcall; +my $decoder = $cdec; +my $lines_per_mapper = 30; +my $iteration = 1; +my $best_weights; +my $psi = 1; +my $default_max_iter = 30; +my $max_iterations = $default_max_iter; +my $jobs = $default_jobs; # number of decode nodes +my $pmem = "4g"; +my $disable_clean = 0; +my %seen_weights; +my $help = 0; +my $epsilon = 0.0001; +my $dryrun = 0; +my $last_score = -10000000; +my $metric = "ibm_bleu"; +my $dir; +my $iniFile; +my $weights; +my $use_make = 1; # use make to parallelize +my $useqsub = 0; +my $initial_weights; +my $pass_suffix = ''; +my $cpbin=1; + +# regularization strength +my $tune_regularizer = 0; +my $reg = 500; +my $reg_previous = 5000; + +# Process command-line options +Getopt::Long::Configure("no_auto_abbrev"); +if (GetOptions( + "jobs=i" => \$jobs, + "dont-clean" => \$disable_clean, + "pass-suffix=s" => \$pass_suffix, + "qsub" => \$useqsub, + "dry-run" => \$dryrun, + "epsilon=s" => \$epsilon, + "help" => \$help, + "weights=s" => \$initial_weights, + "reg=f" => \$reg, + "use-make=i" => \$use_make, + "max-iterations=i" => \$max_iterations, + "pmem=s" => \$pmem, + "cpbin!" => \$cpbin, + "ref-files=s" => \$refFiles, + "metric=s" => \$metric, + "source-file=s" => \$srcFile, + "workdir=s" => \$dir, +) == 0 || @ARGV!=1 || $help) { + print_help(); + exit; +} + +die "--tune-regularizer is no longer supported with --reg-previous and --reg. Please tune manually.\n" if $tune_regularizer; + +if ($useqsub) { + $use_make = 0; + die "LocalEnvironment.pm does not have qsub configuration for this host. Cannot run with --qsub!\n" unless has_qsub(); +} + +my @missing_args = (); +if (!defined $srcFile) { push @missing_args, "--source-file"; } +if (!defined $refFiles) { push @missing_args, "--ref-files"; } +if (!defined $initial_weights) { push @missing_args, "--weights"; } +die "Please specify missing arguments: " . join (', ', @missing_args) . "\n" if (@missing_args); + +if ($metric =~ /^(combi|ter)$/i) { + $lines_per_mapper = 5; +} + +($iniFile) = @ARGV; + + +sub write_config; +sub enseg; +sub print_help; + +my $nodelist; +my $host =check_output("hostname"); chomp $host; +my $bleu; +my $interval_count = 0; +my $logfile; +my $projected_score; + +# used in sorting scores +my $DIR_FLAG = '-r'; +if ($metric =~ /^ter$|^aer$/i) { + $DIR_FLAG = ''; +} + +my $refs_comma_sep = get_comma_sep_refs('r',$refFiles); + +unless ($dir){ + $dir = "rampion"; +} +unless ($dir =~ /^\//){ # convert relative path to absolute path + my $basedir = check_output("pwd"); + chomp $basedir; + $dir = "$basedir/$dir"; +} + + +# Initializations and helper functions +srand; + +my @childpids = (); +my @cleanupcmds = (); + +sub cleanup { + print STDERR "Cleanup...\n"; + for my $pid (@childpids){ unchecked_call("kill $pid"); } + for my $cmd (@cleanupcmds){ unchecked_call("$cmd"); } + exit 1; +}; +# Always call cleanup, no matter how we exit +*CORE::GLOBAL::exit = + sub{ cleanup(); }; +$SIG{INT} = "cleanup"; +$SIG{TERM} = "cleanup"; +$SIG{HUP} = "cleanup"; + +my $decoderBase = check_output("basename $decoder"); chomp $decoderBase; +my $newIniFile = "$dir/$decoderBase.ini"; +my $inputFileName = "$dir/input"; +my $user = $ENV{"USER"}; + + +# process ini file +-e $iniFile || die "Error: could not open $iniFile for reading\n"; +open(INI, $iniFile); + +use File::Basename qw(basename); +#pass bindir, refs to vars holding bin +sub modbin { + local $_; + my $bindir=shift; + check_call("mkdir -p $bindir"); + -d $bindir || die "couldn't make bindir $bindir"; + for (@_) { + my $src=$$_; + $$_="$bindir/".basename($src); + check_call("cp -p $src $$_"); + } +} +sub dirsize { + opendir ISEMPTY,$_[0]; + return scalar(readdir(ISEMPTY))-1; +} +my @allweights; +if ($dryrun){ + write_config(*STDERR); + exit 0; +} else { + if (-e $dir && dirsize($dir)>1 && -e "$dir/hgs" ){ # allow preexisting logfile, binaries, but not dist-pro.pl outputs + die "ERROR: working dir $dir already exists\n\n"; + } else { + -e $dir || mkdir $dir; + mkdir "$dir/hgs"; + modbin("$dir/bin",\$LocalConfig,\$cdec,\$SCORER,\$MAPINPUT,\$MAPPER,\$parallelize,\$sentserver,\$sentclient,\$libcall) if $cpbin; + mkdir "$dir/scripts"; + my $cmdfile="$dir/rerun-pro.sh"; + open CMD,'>',$cmdfile; + print CMD "cd ",&getcwd,"\n"; +# print CMD &escaped_cmdline,"\n"; #buggy - last arg is quoted. + my $cline=&cmdline."\n"; + print CMD $cline; + close CMD; + print STDERR $cline; + chmod(0755,$cmdfile); + check_call("cp $initial_weights $dir/weights.0"); + die "Can't find weights.0" unless (-e "$dir/weights.0"); + } + write_config(*STDERR); +} + + +# Generate initial files and values +check_call("cp $iniFile $newIniFile"); +$iniFile = $newIniFile; + +my $newsrc = "$dir/dev.input"; +enseg($srcFile, $newsrc); +$srcFile = $newsrc; +my $devSize = 0; +open F, "<$srcFile" or die "Can't read $srcFile: $!"; +while() { $devSize++; } +close F; + +unless($best_weights){ $best_weights = $weights; } +unless($projected_score){ $projected_score = 0.0; } +$seen_weights{$weights} = 1; + +my $random_seed = int(time / 1000); +my $lastWeightsFile; +my $lastPScore = 0; +# main optimization loop +while (1){ + print STDERR "\n\nITERATION $iteration\n==========\n"; + + if ($iteration > $max_iterations){ + print STDERR "\nREACHED STOPPING CRITERION: Maximum iterations\n"; + last; + } + # iteration-specific files + my $runFile="$dir/run.raw.$iteration"; + my $onebestFile="$dir/1best.$iteration"; + my $logdir="$dir/logs.$iteration"; + my $decoderLog="$logdir/decoder.sentserver.log.$iteration"; + my $scorerLog="$logdir/scorer.log.$iteration"; + check_call("mkdir -p $logdir"); + + + #decode + print STDERR "RUNNING DECODER AT "; + print STDERR unchecked_output("date"); + my $im1 = $iteration - 1; + my $weightsFile="$dir/weights.$im1"; + push @allweights, "-w $dir/weights.$im1"; + `rm -f $dir/hgs/*.gz`; + my $decoder_cmd = "$decoder -c $iniFile --weights$pass_suffix $weightsFile -O $dir/hgs"; + my $pcmd; + if ($use_make) { + $pcmd = "cat $srcFile | $parallelize --use-fork -p $pmem -e $logdir -j $jobs --"; + } else { + $pcmd = "cat $srcFile | $parallelize -p $pmem -e $logdir -j $jobs --"; + } + my $cmd = "$pcmd $decoder_cmd 2> $decoderLog 1> $runFile"; + print STDERR "COMMAND:\n$cmd\n"; + check_bash_call($cmd); + my $num_hgs; + my $num_topbest; + my $retries = 0; + while($retries < 5) { + $num_hgs = check_output("ls $dir/hgs/*.gz | wc -l"); + $num_topbest = check_output("wc -l < $runFile"); + print STDERR "NUMBER OF HGs: $num_hgs\n"; + print STDERR "NUMBER OF TOP-BEST HYPs: $num_topbest\n"; + if($devSize == $num_hgs && $devSize == $num_topbest) { + last; + } else { + print STDERR "Incorrect number of hypergraphs or topbest. Waiting for distributed filesystem and retrying...\n"; + sleep(3); + } + $retries++; + } + die "Dev set contains $devSize sentences, but we don't have topbest and hypergraphs for all these! Decoder failure? Check $decoderLog\n" if ($devSize != $num_hgs || $devSize != $num_topbest); + my $dec_score = check_output("cat $runFile | $SCORER $refs_comma_sep -m $metric"); + chomp $dec_score; + print STDERR "DECODER SCORE: $dec_score\n"; + + # save space + check_call("gzip -f $runFile"); + check_call("gzip -f $decoderLog"); + + # run optimizer + print STDERR "RUNNING OPTIMIZER AT "; + print STDERR unchecked_output("date"); + print STDERR " - GENERATE TRAINING EXEMPLARS\n"; + my $mergeLog="$logdir/prune-merge.log.$iteration"; + + my $score = 0; + my $icc = 0; + my $inweights="$dir/weights.$im1"; + $cmd="$MAPINPUT $dir/hgs > $dir/agenda.$im1"; + print STDERR "COMMAND:\n$cmd\n"; + check_call($cmd); + die "PLEASE IMPL"; + $iteration++; + print STDERR "\n==========\n"; +} + +print STDERR "\nFINAL WEIGHTS: $lastWeightsFile\n(Use -w with the decoder)\n\n"; + +print STDOUT "$lastWeightsFile\n"; + +exit 0; + +sub get_lines { + my $fn = shift @_; + open FL, "<$fn" or die "Couldn't read $fn: $!"; + my $lc = 0; + while() { $lc++; } + return $lc; +} + +sub get_comma_sep_refs { + my ($r,$p) = @_; + my $o = check_output("echo $p"); + chomp $o; + my @files = split /\s+/, $o; + return "-$r " . join(" -$r ", @files); +} + +sub read_weights_file { + my ($file) = @_; + open F, "<$file" or die "Couldn't read $file: $!"; + my @r = (); + my $pm = -1; + while() { + next if /^#/; + next if /^\s*$/; + chomp; + if (/^(.+)\s+(.+)$/) { + my $m = $1; + my $w = $2; + die "Weights out of order: $m <= $pm" unless $m > $pm; + push @r, $w; + } else { + warn "Unexpected feature name in weight file: $_"; + } + } + close F; + return join ' ', @r; +} + +# subs +sub write_config { + my $fh = shift; + my $cleanup = "yes"; + if ($disable_clean) {$cleanup = "no";} + + print $fh "\n"; + print $fh "DECODER: $decoder\n"; + print $fh "INI FILE: $iniFile\n"; + print $fh "WORKING DIR: $dir\n"; + print $fh "SOURCE (DEV): $srcFile\n"; + print $fh "REFS (DEV): $refFiles\n"; + print $fh "EVAL METRIC: $metric\n"; + print $fh "MAX ITERATIONS: $max_iterations\n"; + print $fh "JOBS: $jobs\n"; + print $fh "HEAD NODE: $host\n"; + print $fh "PMEM (DECODING): $pmem\n"; + print $fh "CLEANUP: $cleanup\n"; +} + +sub update_weights_file { + my ($neww, $rfn, $rpts) = @_; + my @feats = @$rfn; + my @pts = @$rpts; + my $num_feats = scalar @feats; + my $num_pts = scalar @pts; + die "$num_feats (num_feats) != $num_pts (num_pts)" unless $num_feats == $num_pts; + open G, ">$neww" or die; + for (my $i = 0; $i < $num_feats; $i++) { + my $f = $feats[$i]; + my $lambda = $pts[$i]; + print G "$f $lambda\n"; + } + close G; +} + +sub enseg { + my $src = shift; + my $newsrc = shift; + open(SRC, $src); + open(NEWSRC, ">$newsrc"); + my $i=0; + while (my $line=){ + chomp $line; + if ($line =~ /^\s* tags, you must include a zero-based id attribute"; + } + } else { + print NEWSRC "$line\n"; + } + $i++; + } + close SRC; + close NEWSRC; + die "Empty dev set!" if ($i == 0); +} + +sub print_help { + + my $executable = check_output("basename $0"); chomp $executable; + print << "Help"; + +Usage: $executable [options] + + $executable [options] + Runs a complete PRO optimization using the ini file specified. + +Required: + + --ref-files + Dev set ref files. This option takes only a single string argument. + To use multiple files (including file globbing), this argument should + be quoted. + + --source-file + Dev set source file. + + --weights + Initial weights file (use empty file to start from 0) + +General options: + + --help + Print this message and exit. + + --max-iterations + Maximum number of iterations to run. If not specified, defaults + to $default_max_iter. + + --metric + Metric to optimize. + Example values: IBM_BLEU, NIST_BLEU, Koehn_BLEU, TER, Combi + + --pass-suffix + If the decoder is doing multi-pass decoding, the pass suffix "2", + "3", etc., is used to control what iteration of weights is set. + + --workdir + Directory for intermediate and output files. If not specified, the + name is derived from the ini filename. Assuming that the ini + filename begins with the decoder name and ends with ini, the default + name of the working directory is inferred from the middle part of + the filename. E.g. an ini file named decoder.foo.ini would have + a default working directory name foo. + +Regularization options: + + --reg + l2 regularization strength [default=500]. The greater this value, + the closer to zero the weights will be. + +Job control options: + + --jobs + Number of decoder processes to run in parallel. [default=$default_jobs] + + --qsub + Use qsub to run jobs in parallel (qsub must be configured in + environment/LocalEnvironment.pm) + + --pmem + Amount of physical memory requested for parallel decoding jobs + (used with qsub requests only) + +Help +} + +sub convert { + my ($str) = @_; + my @ps = split /;/, $str; + my %dict = (); + for my $p (@ps) { + my ($k, $v) = split /=/, $p; + $dict{$k} = $v; + } + return %dict; +} + + +sub cmdline { + return join ' ',($0,@ORIG_ARGV); +} + +#buggy: last arg gets quoted sometimes? +my $is_shell_special=qr{[ \t\n\\><|&;"'`~*?{}$!()]}; +my $shell_escape_in_quote=qr{[\\"\$`!]}; + +sub escape_shell { + my ($arg)=@_; + return undef unless defined $arg; + if ($arg =~ /$is_shell_special/) { + $arg =~ s/($shell_escape_in_quote)/\\$1/g; + return "\"$arg\""; + } + return $arg; +} + +sub escaped_shell_args { + return map {local $_=$_;chomp;escape_shell($_)} @_; +} + +sub escaped_shell_args_str { + return join ' ',&escaped_shell_args(@_); +} + +sub escaped_cmdline { + return "$0 ".&escaped_shell_args_str(@ORIG_ARGV); +} diff --git a/rampion/rampion_cccp.cc b/rampion/rampion_cccp.cc new file mode 100644 index 00000000..6eb3ccf3 --- /dev/null +++ b/rampion/rampion_cccp.cc @@ -0,0 +1,157 @@ +#include +#include +#include +#include + +#include +#include + +#include "filelib.h" +#include "stringlib.h" +#include "weights.h" +#include "hg_io.h" +#include "kbest.h" +#include "viterbi.h" +#include "ns.h" +#include "ns_docscorer.h" + +using namespace std; +namespace po = boost::program_options; + +void InitCommandLine(int argc, char** argv, po::variables_map* conf) { + po::options_description opts("Configuration options"); + opts.add_options() + ("reference,r",po::value >(), "[REQD] Reference translation (tokenized text)") + ("weights,w",po::value(), "[REQD] Weights files from current iterations") + ("input,i",po::value()->default_value("-"), "Input file to map (- is STDIN)") + ("evaluation_metric,m",po::value()->default_value("IBM_BLEU"), "Evaluation metric (ibm_bleu, koehn_bleu, nist_bleu, ter, meteor, etc.)") + ("kbest_size,k",po::value()->default_value(500u), "Top k-hypotheses to extract") + ("cccp_iterations,I", po::value()->default_value(10u), "CCCP iterations (T')") + ("ssd_iterations,J", po::value()->default_value(5u), "Stochastic subgradient iterations (T'')") + ("eta", po::value()->default_value(1e-4), "Step size") + ("regularization_strength,C", po::value()->default_value(1.0), "L2 regularization strength") + ("alpha,a", po::value()->default_value(10.0), "Cost scale (alpha); alpha * [1-metric(y,y')]") + ("help,h", "Help"); + po::options_description dcmdline_options; + dcmdline_options.add(opts); + po::store(parse_command_line(argc, argv, dcmdline_options), *conf); + bool flag = false; + if (!conf->count("reference")) { + cerr << "Please specify one or more references using -r \n"; + flag = true; + } + if (!conf->count("weights")) { + cerr << "Please specify weights using -w \n"; + flag = true; + } + if (flag || conf->count("help")) { + cerr << dcmdline_options << endl; + exit(1); + } +} + +struct HypInfo { + HypInfo() : g(-100.0f) {} + HypInfo(const vector& h, + const SparseVector& feats, + const SegmentEvaluator& scorer, const EvaluationMetric* metric) : hyp(h), x(feats) { + SufficientStats ss; + scorer.Evaluate(hyp, &ss); + g = metric->ComputeScore(ss); + } + + vector hyp; + float g; + SparseVector x; +}; + +void CostAugmentedSearch(const vector& kbest, + const SparseVector& w, + double alpha, + SparseVector* fmap) { + unsigned best_i = 0; + double best = -numeric_limits::infinity(); + for (unsigned i = 0; i < kbest.size(); ++i) { + double s = kbest[i].x.dot(w) + alpha * kbest[i].g; + if (s > best) { + best = s; + best_i = i; + } + } + *fmap = kbest[best_i].x; +} + +// runs lines 4--15 of rampion algorithm +int main(int argc, char** argv) { + po::variables_map conf; + InitCommandLine(argc, argv, &conf); + const string evaluation_metric = conf["evaluation_metric"].as(); + + EvaluationMetric* metric = EvaluationMetric::Instance(evaluation_metric); + DocumentScorer ds(metric, conf["reference"].as >()); + cerr << "Loaded " << ds.size() << " references for scoring with " << evaluation_metric << endl; + double goodsign = 1; + if (metric->IsErrorMetric()) goodsign = -goodsign; + double badsign = -goodsign; + + Hypergraph hg; + string last_file; + ReadFile in_read(conf["input"].as()); + istream &in=*in_read.stream(); + const unsigned kbest_size = conf["kbest_size"].as(); + const unsigned tp = conf["cccp_iterations"].as(); + const unsigned tpp = conf["ssd_iterations"].as(); + const double eta = conf["eta"].as(); + const double reg = conf["regularization_strength"].as(); + const double alpha = conf["alpha"].as(); + SparseVector weights; + { + vector vweights; + const string weightsf = conf["weights"].as(); + Weights::InitFromFile(weightsf, &vweights); + Weights::InitSparseVector(vweights, &weights); + } + string line, file; + vector > kis; + cerr << "Loading hypergraphs...\n"; + while(getline(in, line)) { + istringstream is(line); + int sent_id; + kis.resize(kis.size() + 1); + vector& curkbest = kis.back(); + is >> file >> sent_id; + ReadFile rf(file); + HypergraphIO::ReadFromJSON(rf.stream(), &hg); + hg.Reweight(weights); + KBest::KBestDerivations, ESentenceTraversal> kbest(hg, kbest_size); + + for (int i = 0; i < kbest_size; ++i) { + const KBest::KBestDerivations, ESentenceTraversal>::Derivation* d = + kbest.LazyKthBest(hg.nodes_.size() - 1, i); + if (!d) break; + curkbest.push_back(HypInfo(d->yield, d->feature_values, *ds[sent_id], metric)); + } + } + + cerr << "Hypergraphs loaded.\n"; + vector > goals(kis.size()); // f(x_i,y+,h+) + SparseVector fear; // f(x,y-,h-) + for (unsigned iterp = 1; iterp <= tp; ++iterp) { + cerr << "CCCP Iteration " << iterp << endl; + for (int i = 0; i < goals.size(); ++i) + CostAugmentedSearch(kis[i], weights, goodsign * alpha, &goals[i]); + for (unsigned iterpp = 1; iterpp <= tpp; ++iterpp) { + cerr << " SSD Iteration " << iterpp << endl; + for (int i = 0; i < goals.size(); ++i) { + CostAugmentedSearch(kis[i], weights, badsign * alpha, &fear); + weights -= weights * (eta * reg / goals.size()); + weights += (goals[i] - fear) * eta; + } + } + } + vector w; + weights.init_vector(&w); + Weights::WriteToFile("-", w); + return 0; +} + diff --git a/rampion/rampion_generate_input.pl b/rampion/rampion_generate_input.pl new file mode 100755 index 00000000..b30fc4fd --- /dev/null +++ b/rampion/rampion_generate_input.pl @@ -0,0 +1,18 @@ +#!/usr/bin/perl -w +use strict; + +die "Usage: $0 HG_DIR\n" unless scalar @ARGV == 1; +my $d = shift @ARGV; +die "Can't find directory $d" unless -d $d; + +opendir(DIR, $d) or die "Can't read $d: $!"; +my @hgs = grep { /\.gz$/ } readdir(DIR); +closedir DIR; + +for my $hg (@hgs) { + my $file = $hg; + my $id = $hg; + $id =~ s/(\.json)?\.gz//; + print "$d/$file $id\n"; +} + -- cgit v1.2.3