configure number of translations to keep

git-svn-id: https://ws10smt.googlecode.com/svn/trunk@410 ec762483-ff6d-05da-a07a-a48fb63a330f

1 files changed, 6 insertions, 3 deletions

diff --git a/gi/pipeline/evaluation-pipeline.pl b/gi/pipeline/evaluation-pipeline.pl
index e940a5b9..675fd8c2 100755
--- a/gi/pipeline/evaluation-pipeline.pl
+++ b/gi/pipeline/evaluation-pipeline.pl
@@ -8,6 +8,7 @@ my $SCRIPT_DIR; BEGIN { use Cwd qw/ abs_path /; use File::Basename; $SCRIPT_DIR
 
 my $JOBS = 15;
 my $PMEM = "9G";
+my $NUM_TRANSLATIONS = 30;
 
 # 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
@@ -132,6 +133,7 @@ if (GetOptions(
         "oov-grammar=s" => \$oovgram,
         "data=s" => \$dataDir,
         "pmem=s" => \$PMEM,
+        "n=i" => \$NUM_TRANSLATIONS,
         "features=s@" => \@features,
         "use-fork" => \$usefork,
         "jobs=i" => \$JOBS,
@@ -279,7 +281,7 @@ sub filter {
     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 $out2 $outgrammar";
+    $cmd = "$FILTERBYF $NUM_TRANSLATIONS $out2 $outgrammar";
     safesystem($outgrammar, $cmd) or die "Secondary filtering failed";
   }
   return $outgrammar;
@@ -317,11 +319,12 @@ EOT
 sub print_help {
   print STDERR<<EOT;
 
-Usage: $0 [-c data-config-file] language-pair grammar.bidir.gz [OPTIONS]
+Usage: $0 [-c data-config-file] [-n N] language-pair grammar.bidir.gz [OPTIONS]
 
 Given an induced grammar for an entire corpus (i.e., generated by
 local-gi-pipeline.pl), filter and featurize it for a dev and test set,
-run MERT, report scores.
+run MERT, report scores. Use -n to specify the number of translations
+to keep for a given source (30 is default).
 
 EOT
 }