From 40c7d43f8cd5ccaef22f098937b985d4101368f9 Mon Sep 17 00:00:00 2001
From: Patrick Simianer <simianer@cl.uni-heidelberg.de>
Date: Tue, 27 Sep 2011 00:10:48 +0200
Subject: some updates for hadoop streaming

---
 dtrain/dtrain.cc                          |   2 +-
 dtrain/dtrain.h                           |   2 +-
 dtrain/hstreaming/avgweights.rb           |  27 ---------------------------
 dtrain/hstreaming/cdec.ini                |   4 ++--
 dtrain/hstreaming/dtrain.ini              |   8 +++++---
 dtrain/hstreaming/hadoop-streaming-job.sh |  11 +++++------
 dtrain/hstreaming/nc-wmt11.en.srilm.3.gz  | Bin 0 -> 12173238 bytes
 dtrain/hstreaming/red-avg.rb              |  26 ++++++++++++++++++++++++++
 8 files changed, 40 insertions(+), 40 deletions(-)
 delete mode 100755 dtrain/hstreaming/avgweights.rb
 create mode 100644 dtrain/hstreaming/nc-wmt11.en.srilm.3.gz
 create mode 100755 dtrain/hstreaming/red-avg.rb

diff --git a/dtrain/dtrain.cc b/dtrain/dtrain.cc
index 622cd01e..9969a070 100644
--- a/dtrain/dtrain.cc
+++ b/dtrain/dtrain.cc
@@ -337,7 +337,7 @@ main(int argc, char** argv)
 
     ++ii;
 
-    if (hstreaming) cerr << "reporter:counter:dtrain,sid," << in_split[0] << endl;
+    if (hstreaming) cerr << "reporter:counter:dtrain,sid," << ii << endl;
 
   } // input loop
 
diff --git a/dtrain/dtrain.h b/dtrain/dtrain.h
index 0c27167d..71dfbc7f 100644
--- a/dtrain/dtrain.h
+++ b/dtrain/dtrain.h
@@ -11,7 +11,7 @@
 #include "pairsampling.h"
 
 #define DTRAIN_DOTS 100 // when to display a '.'
-#define DTRAIN_TMP_DIR "/tmp"
+#define DTRAIN_TMP_DIR "/var/hadoop/mapred/local"
 #define DTRAIN_GRAMMAR_DELIM "########EOS########"
 
 using namespace std;
diff --git a/dtrain/hstreaming/avgweights.rb b/dtrain/hstreaming/avgweights.rb
deleted file mode 100755
index d5cfaa4d..00000000
--- a/dtrain/hstreaming/avgweights.rb
+++ /dev/null
@@ -1,27 +0,0 @@
-#!/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/hstreaming/cdec.ini b/dtrain/hstreaming/cdec.ini
index cc0586d6..bea54afe 100644
--- a/dtrain/hstreaming/cdec.ini
+++ b/dtrain/hstreaming/cdec.ini
@@ -1,7 +1,7 @@
 formalism=scfg
 add_pass_through_rules=true
-feature_function=WordPenalty
 cubepruning_pop_limit=30
+scfg_max_span_limit=15
+feature_function=WordPenalty
 feature_function=KLanguageModel nc-wmt11.en.srilm.3.gz
 feature_function=RuleIdentityFeatures
-scfg_max_span_limit=15
diff --git a/dtrain/hstreaming/dtrain.ini b/dtrain/hstreaming/dtrain.ini
index c7ce7f44..14195bf6 100644
--- a/dtrain/hstreaming/dtrain.ini
+++ b/dtrain/hstreaming/dtrain.ini
@@ -1,7 +1,9 @@
 decoder_config=cdec.ini
-kbest=100
-ngrams=4
+k=100
+N=4
 epochs=10
 input=-
-scorer=stupid_bleu
 output=-
+scorer=stupid_bleu
+sample_from=forest
+pair_sampling=rand
diff --git a/dtrain/hstreaming/hadoop-streaming-job.sh b/dtrain/hstreaming/hadoop-streaming-job.sh
index 2cf3f50a..9da0a6c3 100755
--- a/dtrain/hstreaming/hadoop-streaming-job.sh
+++ b/dtrain/hstreaming/hadoop-streaming-job.sh
@@ -5,16 +5,15 @@ 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
+OUT=out/nc-wmt11-de-en-dyer-cs-joshua.tok.lc.fixamp1.loo.psg.dtrain-weights-1400m-NEW
 
 $HSTREAMING \
-    -mapper "dtrain.sh" \
-    -reducer "avgweights.rb" \
+    -mapper "dtrain -c dtrain.ini --hstreaming" \
+    -reducer "red-avg.rb" \
     -input $IN \
     -output $OUT \
-    -file avgweights.rb \
-    -file dtrain.sh \
-    -file dtrain \
+    -file red-avg.rb \
+    -file ../dtrain \
     -file dtrain.ini \
     -file cdec.ini \
     -file nc-wmt11.en.srilm.3.gz \
diff --git a/dtrain/hstreaming/nc-wmt11.en.srilm.3.gz b/dtrain/hstreaming/nc-wmt11.en.srilm.3.gz
new file mode 100644
index 00000000..5a50f8fb
Binary files /dev/null and b/dtrain/hstreaming/nc-wmt11.en.srilm.3.gz differ
diff --git a/dtrain/hstreaming/red-avg.rb b/dtrain/hstreaming/red-avg.rb
new file mode 100755
index 00000000..11dc0d71
--- /dev/null
+++ b/dtrain/hstreaming/red-avg.rb
@@ -0,0 +1,26 @@
+#!/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
+  puts "#{k}\t{w[k]/shard_count}"
+}
+
+puts "#{shard_count_key}\t#{w[shard_count_key]}"
+
-- 
cgit v1.2.3