latest version from mtm6

17 files changed, 218 insertions, 11 deletions

diff --git a/dtrain/avgweights.rb b/dtrain/avgweights.rb
new file mode 100755
index 00000000..d5cfaa4d
--- /dev/null
+++ b/dtrain/avgweights.rb
@@ -0,0 +1,27 @@
+#!/usr/bin/env ruby1.9.1
+
+
+STDIN.set_encoding 'utf-8'
+
+#shard_count_key = "__SHARD_COUNT__"
+
+w = {}
+c = {}
+w.default = 0
+c.default = 0
+while line = STDIN.gets
+  key, val = line.split /\t/
+  w[key] += val.to_f
+  c[key] += 1.0
+end
+
+#shard_count = w["__SHARD_COUNT__"]
+
+w.each_key { |k|
+  #if k == shard_count_key then next end
+  #if k == "__bias" then next end
+  puts "#{k}\t#{w[k]/c[k]}" #{w[k]/shard_count}"
+}
+
+#puts "#{shard_count_key}\t#{w[shard_count_key]}"
+
diff --git a/dtrain/common.h b/dtrain/common.h
index 4ff975e1..49dc85b7 100644
--- a/dtrain/common.h
+++ b/dtrain/common.h
@@ -30,7 +30,7 @@
 #define DTRAIN_DEFAULT_T 1                  // iterations
 #define DTRAIN_DEFAULT_SCORER "stupid_bleu" // scorer
 #define DTRAIN_DOTS 100                     // when to display a '.'
-#define DTRAIN_TMP_DIR "/tmp"               // put this on a SSD?
+#define DTRAIN_TMP_DIR "/var/hadoop/mapred/local"               // put this on a SSD?
 #define DTRAIN_GRAMMAR_DELIM "########EOS########"
 
 
diff --git a/dtrain/dtrain.cc b/dtrain/dtrain.cc
index 35996d6d..f005008e 100644
--- a/dtrain/dtrain.cc
+++ b/dtrain/dtrain.cc
@@ -13,6 +13,8 @@
 //#include <boost/iostreams/filter/bzip2.hpp>
 using namespace boost::iostreams;
 
+#include <boost/algorithm/string/predicate.hpp>
+#include <boost/lexical_cast.hpp>
 
 #ifdef DTRAIN_DEBUG
 #include "tests.h"
@@ -311,7 +313,7 @@ main( int argc, char** argv )
       }
       if ( broken_grammar ) continue;
       grammar_str = boost::replace_all_copy( in_split[3], " __NEXT__RULE__ ", "\n" ) + "\n"; // FIXME copy, __
-      grammar_buf << grammar_str << DTRAIN_GRAMMAR_DELIM << endl;
+      grammar_buf << grammar_str << DTRAIN_GRAMMAR_DELIM << " " << in_split[0] << endl;
       decoder.SetSentenceGrammarFromString( grammar_str );
       // decode, kbest
       src_str_buf.push_back( in_split[1] );
@@ -323,7 +325,8 @@ main( int argc, char** argv )
       while ( true ) {
         string g;  
         getline( grammar_buf_in, g );
-        if ( g == DTRAIN_GRAMMAR_DELIM ) break;
+        //if ( g == DTRAIN_GRAMMAR_DELIM ) break;
+        if (boost::starts_with(g, DTRAIN_GRAMMAR_DELIM)) break;
         grammar_str += g+"\n";
         i += 1;
       }
@@ -430,6 +433,7 @@ main( int argc, char** argv )
     }
 
     ++sid;
+    cerr << "reporter:counter:dtrain,sent," << sid << endl;
 
   } // input loop
 
@@ -446,6 +450,7 @@ main( int argc, char** argv )
     avg_1best_score_diff = avg_1best_score;
     avg_1best_model_diff = avg_1best_model;
   }
+  if ( !quiet ) {
   cout << _prec5 << _pos << "WEIGHTS" << endl;
   for (vector<string>::iterator it = wprint.begin(); it != wprint.end(); it++) {
     cout << setw(16) << *it << " = " << dense_weights[FD::Convert( *it )] << endl;
@@ -456,6 +461,7 @@ main( int argc, char** argv )
   cout << _pos << " (" << avg_1best_score_diff << ")" << endl;
   cout << _nopos << "avg model score: " << avg_1best_model;
   cout << _pos << " (" << avg_1best_model_diff << ")" << endl;
+  }
   vector<double> remember_scores;
   remember_scores.push_back( avg_1best_score );
   remember_scores.push_back( avg_1best_model );
@@ -478,7 +484,7 @@ main( int argc, char** argv )
     cout << time_dif/(double)in_sz<< " s/S)" << endl;
   }
   
-  if ( t+1 != T ) cout << endl;
+  if ( t+1 != T && !quiet ) cout << endl;
 
   if ( noup ) break;
 
@@ -486,8 +492,21 @@ main( int argc, char** argv )
 
   unlink( grammar_buf_tmp_fn );
   if ( !noup ) {
+    // TODO BEST ITER
     if ( !quiet ) cout << endl << "writing weights file '" << cfg["output"].as<string>() << "' ...";
-    weights.WriteToFile( cfg["output"].as<string>(), true );
+    if ( cfg["output"].as<string>() == "-" ) {
+      for ( SparseVector<double>::const_iterator ti = lambdas.begin();
+            ti != lambdas.end(); ++ti ) {
+	if ( ti->second == 0 ) continue;
+        //if ( ti->first == "__bias" ) continue;
+        cout << setprecision(9);
+        cout << _nopos << FD::Convert(ti->first) << "\t" << ti->second << endl;
+        //cout << "__SHARD_COUNT__\t1" << endl;
+      }
+    } else {
+      weights.InitFromVector( lambdas );
+      weights.WriteToFile( cfg["output"].as<string>(), true );
+    }
     if ( !quiet ) cout << "done" << endl;
   }
   
diff --git a/dtrain/job/avgweights.rb b/dtrain/job/avgweights.rb
new file mode 100755
index 00000000..e635aab4
--- /dev/null
+++ b/dtrain/job/avgweights.rb
@@ -0,0 +1,30 @@
+#!/usr/bin/env ruby1.9.1
+
+
+STDIN.set_encoding 'utf-8'
+
+#shard_count_key = "__SHARD_COUNT__"
+
+w = {}
+#c = {}
+w.default = 0
+#c.default = 0
+while line = STDIN.gets
+  key, val = line.split /\t/
+  w[key] += val.to_f
+  #c[key] += 1.0
+end
+
+#shard_count = w["__SHARD_COUNT__"]
+
+num_map = 104.0
+
+w.each_key { |k|
+  #if k == shard_count_key then next end
+  #if k == "__bias" then next end
+  puts "#{k}\t#{w[k]/num_map}"
+  #/c[k]}" #{w[k]/shard_count}"
+}
+
+#puts "#{shard_count_key}\t#{w[shard_count_key]}"
+
diff --git a/dtrain/job/cdec.ini b/dtrain/job/cdec.ini
new file mode 100644
index 00000000..0d32f0b7
--- /dev/null
+++ b/dtrain/job/cdec.ini
@@ -0,0 +1,8 @@
+formalism=scfg
+add_pass_through_rules=true
+feature_function=WordPenalty
+cubepruning_pop_limit=30
+feature_function=KLanguageModel nc-wmt11.en.srilm.3.gz
+feature_function=RuleIdentityFeatures
+scfg_max_span_limit=15
+
diff --git a/dtrain/job/dtrain.ini b/dtrain/job/dtrain.ini
new file mode 100644
index 00000000..079d7d69
--- /dev/null
+++ b/dtrain/job/dtrain.ini
@@ -0,0 +1,10 @@
+decoder_config=cdec.ini
+kbest=100
+ngrams=4
+epochs=10
+input=-
+scorer=stupid_bleu
+output=-
+#stop_after=100
+#wprint=Glue WordPenalty LanguageModel LanguageModel_OOV PhraseModel_0 PhraseModel_1 PhraseModel_2 PhraseModel_3 PhraseModel_4 PassThrough
+
diff --git a/dtrain/job/dtrain.sh b/dtrain/job/dtrain.sh
new file mode 100755
index 00000000..75ec29ea
--- /dev/null
+++ b/dtrain/job/dtrain.sh
@@ -0,0 +1,6 @@
+#!/bin/sh
+
+./dtrain -q -c dtrain.ini
+
+exit 0
+
diff --git a/dtrain/job/hadoop-streaming-job.sh b/dtrain/job/hadoop-streaming-job.sh
new file mode 100755
index 00000000..2cf3f50a
--- /dev/null
+++ b/dtrain/job/hadoop-streaming-job.sh
@@ -0,0 +1,23 @@
+#!/bin/bash
+
+HADOOP_HOME=/usr/lib/hadoop-0.20
+JAR=contrib/streaming/hadoop-streaming-0.20.2-cdh3u1.jar
+HSTREAMING="$HADOOP_HOME/bin/hadoop jar $HADOOP_HOME/$JAR"
+
+IN=in/nc-wmt11-de-en-dyer-cs-joshua.tok.lc.fixamp1.loo.psg.dtrain.1400m
+OUT=out/nc-wmt11-de-en-dyer-cs-joshua.tok.lc.fixamp1.loo.psg.dtrain-weights-1400m
+
+$HSTREAMING \
+    -mapper "dtrain.sh" \
+    -reducer "avgweights.rb" \
+    -input $IN \
+    -output $OUT \
+    -file avgweights.rb \
+    -file dtrain.sh \
+    -file dtrain \
+    -file dtrain.ini \
+    -file cdec.ini \
+    -file nc-wmt11.en.srilm.3.gz \
+    -jobconf mapred.reduce.tasks=1 \
+    -jobconf mapred.max.map.failures.percent=100
+
diff --git a/dtrain/job2/avgweights.rb b/dtrain/job2/avgweights.rb
new file mode 100755
index 00000000..31048f16
--- /dev/null
+++ b/dtrain/job2/avgweights.rb
@@ -0,0 +1,30 @@
+#!/usr/bin/env ruby1.9.1
+
+
+STDIN.set_encoding 'utf-8'
+
+#shard_count_key = "__SHARD_COUNT__"
+
+w = {}
+#c = {}
+w.default = 0
+#c.default = 0
+while line = STDIN.gets
+  key, val = line.split /\t/
+  w[key] += val.to_f
+  #c[key] += 1.0
+end
+
+#shard_count = w["__SHARD_COUNT__"]
+
+num_map = 2107.0
+
+w.each_key { |k|
+  #if k == shard_count_key then next end
+  #if k == "__bias" then next end
+  puts "#{k}\t#{w[k]/num_map}"
+  #/c[k]}" #{w[k]/shard_count}"
+}
+
+#puts "#{shard_count_key}\t#{w[shard_count_key]}"
+
diff --git a/dtrain/job2/cdec.ini b/dtrain/job2/cdec.ini
new file mode 100644
index 00000000..0d32f0b7
--- /dev/null
+++ b/dtrain/job2/cdec.ini
@@ -0,0 +1,8 @@
+formalism=scfg
+add_pass_through_rules=true
+feature_function=WordPenalty
+cubepruning_pop_limit=30
+feature_function=KLanguageModel nc-wmt11.en.srilm.3.gz
+feature_function=RuleIdentityFeatures
+scfg_max_span_limit=15
+
diff --git a/dtrain/job2/dtrain.ini b/dtrain/job2/dtrain.ini
new file mode 100644
index 00000000..ec005e46
--- /dev/null
+++ b/dtrain/job2/dtrain.ini
@@ -0,0 +1,10 @@
+decoder_config=cdec.ini
+kbest=100
+ngrams=3
+epochs=10
+input=-
+scorer=stupid_bleu
+output=-
+#stop_after=100
+#wprint=Glue WordPenalty LanguageModel LanguageModel_OOV PhraseModel_0 PhraseModel_1 PhraseModel_2 PhraseModel_3 PhraseModel_4 PassThrough
+
diff --git a/dtrain/job2/dtrain.sh b/dtrain/job2/dtrain.sh
new file mode 100755
index 00000000..75ec29ea
--- /dev/null
+++ b/dtrain/job2/dtrain.sh
@@ -0,0 +1,6 @@
+#!/bin/sh
+
+./dtrain -q -c dtrain.ini
+
+exit 0
+
diff --git a/dtrain/job2/hadoop-streaming-job.sh b/dtrain/job2/hadoop-streaming-job.sh
new file mode 100755
index 00000000..9ee70a33
--- /dev/null
+++ b/dtrain/job2/hadoop-streaming-job.sh
@@ -0,0 +1,23 @@
+#!/bin/bash
+
+HADOOP_HOME=/usr/lib/hadoop-0.20
+JAR=contrib/streaming/hadoop-streaming-0.20.2-cdh3u1.jar
+HSTREAMING="$HADOOP_HOME/bin/hadoop jar $HADOOP_HOME/$JAR"
+
+IN=in/nc-wmt11-de-en-dyer-cs-joshua.tok.lc.fixamp1.loo.psg.dtrain
+OUT=out/nc-wmt11-de-en-dyer-cs-joshua.tok.lc.fixamp1.loo.psg.dtrain-weights.50
+
+$HSTREAMING \
+    -mapper "dtrain.sh" \
+    -reducer "avgweights.rb" \
+    -input $IN \
+    -output $OUT \
+    -file avgweights.rb \
+    -file dtrain.sh \
+    -file dtrain \
+    -file dtrain.ini \
+    -file cdec.ini \
+    -file nc-wmt11.en.srilm.3.gz \
+    -jobconf mapred.reduce.tasks=1 \
+    -jobconf mapred.max.map.failures.percent=100
+
diff --git a/dtrain/run.sh b/dtrain/run.sh
index 97123dfa..72e56f3e 100755
--- a/dtrain/run.sh
+++ b/dtrain/run.sh
@@ -2,11 +2,11 @@
 
 #INI=test/blunsom08.dtrain.ini
 #INI=test/nc-wmt11/dtrain.ini
-INI=test/EXAMPLE/dtrain.ini
+#INI=test/EXAMPLE/dtrain.ini
 #INI=test/EXAMPLE/dtrain.ruleids.ini
-#INI=test/toy.dtrain.ini
+INI=test/toy.dtrain.ini
 #INI=test/EXAMPLE/dtrain.cdecrid.ini
 
-rm /tmp/dtrain-*
+#rm /tmp/dtrain-*
 ./dtrain -c $INI $1 $2 $3 $4 
 
diff --git a/dtrain/test-reducer b/dtrain/test-reducer
new file mode 100644
index 00000000..b86e7894
--- /dev/null
+++ b/dtrain/test-reducer
@@ -0,0 +1,7 @@
+a	1
+b	2
+c	3.5
+a	1
+b	2
+c	3.5
+__SHARD_COUNT__	2
diff --git a/dtrain/test/toy.dtrain.ini b/dtrain/test/toy.dtrain.ini
index e9ed0ce5..7272e655 100644
--- a/dtrain/test/toy.dtrain.ini
+++ b/dtrain/test/toy.dtrain.ini
@@ -2,9 +2,9 @@ decoder_config=test/cdec.ini
 kbest=4
 ngrams=1
 epochs=3
-input=test/toy.in
+input=- #test/toy.in
 scorer=bleu
-output=data/w/toy.gz
+output=- #data/w/toy.gz
 #stop_after=1000
 wprint=logp use_shell use_house PassThrough
 
diff --git a/utils/sampler.h b/utils/sampler.h
index a14f6e2f..8567e922 100644
--- a/utils/sampler.h
+++ b/utils/sampler.h
@@ -32,7 +32,7 @@ struct RandomNumberGenerator {
       std::cerr << "Warning: could not read from /dev/urandom. Seeding from clock" << std::endl;
       seed = std::time(NULL);
     }
-    std::cerr << "Seeding random number sequence to " << seed << std::endl;
+    //std::cerr << "Seeding random number sequence to " << seed << std::endl;
     return seed;
   }