From 671c21451542e2dd20e45b4033d44d8e8735f87b Mon Sep 17 00:00:00 2001
From: Chris Dyer <redpony@gmail.com>
Date: Thu, 3 Dec 2009 16:33:55 -0500
Subject: initial check in

---
 training/cluster-ptrain.pl | 144 +++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 144 insertions(+)
 create mode 100755 training/cluster-ptrain.pl

(limited to 'training/cluster-ptrain.pl')

diff --git a/training/cluster-ptrain.pl b/training/cluster-ptrain.pl
new file mode 100755
index 00000000..99369cdc
--- /dev/null
+++ b/training/cluster-ptrain.pl
@@ -0,0 +1,144 @@
+#!/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; }
+use Getopt::Long;
+
+my $MAX_ITER_ATTEMPTS = 5; # number of times to retry a failed function evaluation
+my $CWD=`pwd`; chomp $CWD;
+my $BIN_DIR = $SCRIPT_DIR;
+my $OPTIMIZER = "$BIN_DIR/mr_optimize_reduce";
+my $DECODER = "$BIN_DIR/cdec";
+my $COMBINER_CACHE_SIZE = 150;
+my $PARALLEL = "/chomes/redpony/svn-trunk/sa-utils/parallelize.pl";
+die "Can't find $OPTIMIZER" unless -f $OPTIMIZER;
+die "Can't execute $OPTIMIZER" unless -x $OPTIMIZER;
+my $restart = '';
+if ($ARGV[0] && $ARGV[0] eq '--restart') { shift @ARGV; $restart = 1; }
+
+my $pmem="2500mb";
+my $nodes = 36;
+my $max_iteration = 1000;
+my $PRIOR_FLAG = "";
+my $parallel = 1;
+my $CFLAG = "-C 1";
+my $LOCAL;
+my $PRIOR;
+my $OALG = "lbfgs";
+my $sigsq = 1;
+my $means_file;
+GetOptions("decoder=s" => \$DECODER,
+           "run_locally" => \$LOCAL,
+           "gaussian_prior" => \$PRIOR,
+           "sigma_squared=f" => \$sigsq,
+           "means=s" => \$means_file,
+           "optimizer=s" => \$OALG,
+           "pmem=s" => \$pmem
+          ) or usage();
+usage() unless scalar @ARGV==3;
+my $config_file = shift @ARGV;
+my $training_corpus = shift @ARGV;
+my $initial_weights = shift @ARGV;
+die "Can't find $config_file" unless -f $config_file;
+die "Can't find $DECODER" unless -f $DECODER;
+die "Can't execute $DECODER" unless -x $DECODER;
+if ($LOCAL) { print STDERR "Will running LOCALLY.\n"; $parallel = 0; }
+if ($PRIOR) {
+  $PRIOR_FLAG="-p --sigma_squared $sigsq";
+  if ($means_file) { $PRIOR_FLAG .= " -u $means_file"; }
+}
+
+if ($parallel) {
+  die "Can't find $PARALLEL" unless -f $PARALLEL;
+  die "Can't execute $PARALLEL" unless -x $PARALLEL;
+}
+unless ($parallel) { $CFLAG = "-C 500"; }
+unless ($config_file =~ /^\//) { $config_file = $CWD . '/' . $config_file; }
+
+print STDERR <<EOT;
+PTRAIN CONFIGURATION INFORMATION
+
+      Config file: $config_file
+  Training corpus: $training_corpus
+  Initial weights: $initial_weights
+   Decoder memory: $pmem
+  Nodes requested: $nodes
+   Max iterations: $max_iteration
+        Optimizer: $OALG
+            PRIOR: $PRIOR_FLAG
+          restart: $restart
+EOT
+if ($OALG) { $OALG="-m $OALG"; }
+
+my $nodelist="1";
+for (my $i=1; $i<$nodes; $i++) { $nodelist .= " 1"; }
+my $iter = 1;
+
+my $dir = "$CWD/ptrain";
+if ($restart) {
+  die "$dir doesn't exist, but --restart specified!\n" unless -d $dir;
+  my $o = `ls -t $dir/weights.*`;
+  my ($a, @x) = split /\n/, $o;
+  if ($a =~ /weights.(\d+)\.gz$/) {
+    $iter = $1;
+  } else {
+    die "Unexpected file: $a!\n";
+  }
+  print STDERR "Restarting at iteration $iter\n";
+} else {
+  die "$dir already exists!\n" if -e $dir;
+  mkdir $dir or die "Can't create $dir: $!";
+
+  unless ($initial_weights =~ /\.gz$/) {
+    `cp $initial_weights $dir/weights.1`;
+    `gzip -9 $dir/weights.1`;
+  } else {
+    `cp $initial_weights $dir/weights.1.gz`;
+  }
+}
+
+my $iter_attempts = 1;
+while ($iter < $max_iteration) {
+  my $cur_time = `date`; chomp $cur_time;
+  print STDERR "\nStarting iteration $iter...\n";
+  print STDERR "  time: $cur_time\n";
+  my $start = time;
+  my $next_iter = $iter + 1;
+  my $dec_cmd="$DECODER -G $CFLAG -c $config_file -w $dir/weights.$iter.gz < $training_corpus 2> $dir/deco.log.$iter";
+  my $opt_cmd = "$OPTIMIZER $PRIOR_FLAG -M 50 $OALG -s $dir/opt.state -i $dir/weights.$iter.gz -o $dir/weights.$next_iter.gz";
+  my $pcmd = "$PARALLEL -e $dir/err -p $pmem --nodelist \"$nodelist\" -- ";
+  my $cmd = "";
+  if ($parallel) { $cmd = $pcmd; }
+  $cmd .= "$dec_cmd | $opt_cmd";
+
+  print STDERR "EXECUTING: $cmd\n";
+  my $result = `$cmd`;
+  my $exit_code = $? >> 8;
+  if ($exit_code == 99) {
+    $iter_attempts++;
+    if ($iter_attempts > $MAX_ITER_ATTEMPTS) {
+      die "Received restart request $iter_attempts times from optimizer, giving up\n";
+    }
+    print STDERR "Function evaluation failed, retrying (attempt $iter_attempts)\n";
+    next;
+  }
+  if ($? != 0) {
+    die "Error running iteration $iter: $!";
+  }
+  chomp $result;
+  my $end = time;
+  my $diff = ($end - $start);
+  print STDERR "  ITERATION $iter TOOK $diff SECONDS\n";
+  $iter = $next_iter;
+  if ($result =~ /1$/) {
+    print STDERR "Training converged.\n";
+    last;
+  }
+  $iter_attempts = 1;
+}
+
+print "FINAL WEIGHTS: $dir/weights.$iter\n";
+
+sub usage {
+  die "Usage: $0 [OPTIONS] cdec.ini training.corpus weights.init\n";
+}
-- 
cgit v1.2.3