Merge branch 'master' of https://github.com/redpony/cdec

43 files changed, 640 insertions, 0 deletions

diff --git a/training/dtrain/examples/parallelized/README b/training/dtrain/examples/parallelized/README
new file mode 100644
index 00000000..89715105
--- /dev/null
+++ b/training/dtrain/examples/parallelized/README
@@ -0,0 +1,5 @@
+run for example
+  ../../parallelize.rb ./dtrain.ini 4 false 2 2 ./in ./refs
+
+final weights will be in the file work/weights.3
+
diff --git a/training/dtrain/examples/parallelized/cdec.ini b/training/dtrain/examples/parallelized/cdec.ini
new file mode 100644
index 00000000..e43ba1c4
--- /dev/null
+++ b/training/dtrain/examples/parallelized/cdec.ini
@@ -0,0 +1,22 @@
+formalism=scfg
+add_pass_through_rules=true
+intersection_strategy=cube_pruning
+cubepruning_pop_limit=200
+scfg_max_span_limit=15
+feature_function=WordPenalty
+feature_function=KLanguageModel ../example/nc-wmt11.en.srilm.gz
+#feature_function=ArityPenalty
+#feature_function=CMR2008ReorderingFeatures
+#feature_function=Dwarf
+#feature_function=InputIndicator
+#feature_function=LexNullJump
+#feature_function=NewJump
+#feature_function=NgramFeatures
+#feature_function=NonLatinCount
+#feature_function=OutputIndicator
+#feature_function=RuleIdentityFeatures
+#feature_function=RuleNgramFeatures
+#feature_function=RuleShape
+#feature_function=SourceSpanSizeFeatures
+#feature_function=SourceWordPenalty
+#feature_function=SpanFeatures
diff --git a/training/dtrain/examples/parallelized/dtrain.ini b/training/dtrain/examples/parallelized/dtrain.ini
new file mode 100644
index 00000000..f19ef891
--- /dev/null
+++ b/training/dtrain/examples/parallelized/dtrain.ini
@@ -0,0 +1,16 @@
+k=100
+N=4
+learning_rate=0.0001
+gamma=0
+loss_margin=1.0
+epochs=1
+scorer=stupid_bleu
+sample_from=kbest
+filter=uniq
+pair_sampling=XYX
+hi_lo=0.1
+select_weights=last
+print_weights=Glue WordPenalty LanguageModel LanguageModel_OOV PhraseModel_0 PhraseModel_1 PhraseModel_2 PhraseModel_3 PhraseModel_4 PhraseModel_5 PhraseModel_6 PassThrough
+# newer version of the grammar extractor use different feature names: 
+#print_weights=Glue WordPenalty LanguageModel LanguageModel_OOV PhraseModel_0 PhraseModel_1 PhraseModel_2 PhraseModel_3 PhraseModel_4 PhraseModel_5 PhraseModel_6 PassThrough
+decoder_config=cdec.ini
diff --git a/training/dtrain/examples/parallelized/grammar/grammar.out.0.gz b/training/dtrain/examples/parallelized/grammar/grammar.out.0.gz
new file mode 100644
index 00000000..1e28a24b
--- /dev/null
+++ b/training/dtrain/examples/parallelized/grammar/grammar.out.0.gz
diff --git a/training/dtrain/examples/parallelized/grammar/grammar.out.1.gz b/training/dtrain/examples/parallelized/grammar/grammar.out.1.gz
new file mode 100644
index 00000000..372f5675
--- /dev/null
+++ b/training/dtrain/examples/parallelized/grammar/grammar.out.1.gz
diff --git a/training/dtrain/examples/parallelized/grammar/grammar.out.2.gz b/training/dtrain/examples/parallelized/grammar/grammar.out.2.gz
new file mode 100644
index 00000000..145d0dc0
--- /dev/null
+++ b/training/dtrain/examples/parallelized/grammar/grammar.out.2.gz
diff --git a/training/dtrain/examples/parallelized/grammar/grammar.out.3.gz b/training/dtrain/examples/parallelized/grammar/grammar.out.3.gz
new file mode 100644
index 00000000..105593ff
--- /dev/null
+++ b/training/dtrain/examples/parallelized/grammar/grammar.out.3.gz
diff --git a/training/dtrain/examples/parallelized/grammar/grammar.out.4.gz b/training/dtrain/examples/parallelized/grammar/grammar.out.4.gz
new file mode 100644
index 00000000..30781f48
--- /dev/null
+++ b/training/dtrain/examples/parallelized/grammar/grammar.out.4.gz
diff --git a/training/dtrain/examples/parallelized/grammar/grammar.out.5.gz b/training/dtrain/examples/parallelized/grammar/grammar.out.5.gz
new file mode 100644
index 00000000..834ee759
--- /dev/null
+++ b/training/dtrain/examples/parallelized/grammar/grammar.out.5.gz
diff --git a/training/dtrain/examples/parallelized/grammar/grammar.out.6.gz b/training/dtrain/examples/parallelized/grammar/grammar.out.6.gz
new file mode 100644
index 00000000..2e76f348
--- /dev/null
+++ b/training/dtrain/examples/parallelized/grammar/grammar.out.6.gz
diff --git a/training/dtrain/examples/parallelized/grammar/grammar.out.7.gz b/training/dtrain/examples/parallelized/grammar/grammar.out.7.gz
new file mode 100644
index 00000000..3741a887
--- /dev/null
+++ b/training/dtrain/examples/parallelized/grammar/grammar.out.7.gz
diff --git a/training/dtrain/examples/parallelized/grammar/grammar.out.8.gz b/training/dtrain/examples/parallelized/grammar/grammar.out.8.gz
new file mode 100644
index 00000000..ebf6bd0c
--- /dev/null
+++ b/training/dtrain/examples/parallelized/grammar/grammar.out.8.gz
diff --git a/training/dtrain/examples/parallelized/grammar/grammar.out.9.gz b/training/dtrain/examples/parallelized/grammar/grammar.out.9.gz
new file mode 100644
index 00000000..c1791059
--- /dev/null
+++ b/training/dtrain/examples/parallelized/grammar/grammar.out.9.gz
diff --git a/training/dtrain/examples/parallelized/in b/training/dtrain/examples/parallelized/in
new file mode 100644
index 00000000..51d01fe7
--- /dev/null
+++ b/training/dtrain/examples/parallelized/in
@@ -0,0 +1,10 @@
+<seg grammar="grammar/grammar.out.0.gz" id="0">europas nach rassen geteiltes haus</seg>
+<seg grammar="grammar/grammar.out.1.gz" id="1">ein gemeinsames merkmal aller extremen rechten in europa ist ihr rassismus und die tatsache , daß sie das einwanderungsproblem als politischen hebel benutzen .</seg>
+<seg grammar="grammar/grammar.out.2.gz" id="2">der lega nord in italien , der vlaams block in den niederlanden , die anhänger von le pens nationaler front in frankreich , sind beispiele für parteien oder bewegungen , die sich um das gemeinsame thema : ablehnung der zuwanderung gebildet haben und um forderung nach einer vereinfachten politik , um sie zu regeln .</seg>
+<seg grammar="grammar/grammar.out.3.gz" id="3">während individuen wie jörg haidar und jean @-@ marie le pen kommen und ( leider nicht zu bald ) wieder gehen mögen , wird die rassenfrage aus der europäischer politik nicht so bald verschwinden .</seg>
+<seg grammar="grammar/grammar.out.4.gz" id="4">eine alternde einheimische bevölkerung und immer offenere grenzen vermehren die rassistische zersplitterung in den europäischen ländern .</seg>
+<seg grammar="grammar/grammar.out.5.gz" id="5">die großen parteien der rechten und der linken mitte haben sich dem problem gestellt , in dem sie den kopf in den sand gesteckt und allen aussichten zuwider gehofft haben , es möge bald verschwinden .</seg>
+<seg grammar="grammar/grammar.out.6.gz" id="6">das aber wird es nicht , wie die geschichte des rassismus in amerika deutlich zeigt .</seg>
+<seg grammar="grammar/grammar.out.7.gz" id="7">die beziehungen zwischen den rassen standen in den usa über jahrzehnte - und tun das noch heute - im zentrum der politischen debatte . das ging so weit , daß rassentrennung genauso wichtig wie das einkommen wurde , - wenn nicht sogar noch wichtiger - um politische zuneigungen und einstellungen zu bestimmen .</seg>
+<seg grammar="grammar/grammar.out.8.gz" id="8">der erste schritt , um mit der rassenfrage umzugehen ist , ursache und folgen rassistischer feindseligkeiten zu verstehen , auch dann , wenn das bedeutet , unangenehme tatsachen aufzudecken .</seg>
+<seg grammar="grammar/grammar.out.9.gz" id="9">genau das haben in den usa eine große anzahl an forschungsvorhaben in wirtschaft , soziologie , psychologie und politikwissenschaft geleistet . diese forschungen zeigten , daß menschen unterschiedlicher rasse einander deutlich weniger vertrauen .</seg>
diff --git a/training/dtrain/examples/parallelized/refs b/training/dtrain/examples/parallelized/refs
new file mode 100644
index 00000000..632e27b0
--- /dev/null
+++ b/training/dtrain/examples/parallelized/refs
@@ -0,0 +1,10 @@
+europe 's divided racial house
+a common feature of europe 's extreme right is its racism and use of the immigration issue as a political wedge .
+the lega nord in italy , the vlaams blok in the netherlands , the supporters of le pen 's national front in france , are all examples of parties or movements formed on the common theme of aversion to immigrants and promotion of simplistic policies to control them .
+while individuals like jorg haidar and jean @-@ marie le pen may come and ( never to soon ) go , the race question will not disappear from european politics anytime soon .
+an aging population at home and ever more open borders imply increasing racial fragmentation in european countries .
+mainstream parties of the center left and center right have confronted this prospect by hiding their heads in the ground , hoping against hope that the problem will disappear .
+it will not , as america 's racial history clearly shows .
+race relations in the us have been for decades - and remain - at the center of political debate , to the point that racial cleavages are as important as income , if not more , as determinants of political preferences and attitudes .
+the first step to address racial politics is to understand the origin and consequences of racial animosity , even if it means uncovering unpleasant truths .
+this is precisely what a large amount of research in economics , sociology , psychology and political science has done for the us .
diff --git a/training/dtrain/examples/parallelized/work/out.0.0 b/training/dtrain/examples/parallelized/work/out.0.0
new file mode 100644
index 00000000..7a00ed0f
--- /dev/null
+++ b/training/dtrain/examples/parallelized/work/out.0.0
@@ -0,0 +1,61 @@
+                cdec cfg 'cdec.ini'
+Loading the LM will be faster if you build a binary file.
+Reading ../example/nc-wmt11.en.srilm.gz
+----5---10---15---20---25---30---35---40---45---50---55---60---65---70---75---80---85---90---95--100
+****************************************************************************************************
+Seeding random number sequence to 3121929377
+
+dtrain
+Parameters:
+                       k 100
+                       N 4
+                       T 1
+                  scorer 'stupid_bleu'
+             sample from 'kbest'
+                  filter 'uniq'
+           learning rate 0.0001
+                   gamma 0
+             loss margin 1
+                   pairs 'XYX'
+                   hi lo 0.1
+          pair threshold 0
+          select weights 'last'
+                  l1 reg 0 'none'
+               max pairs 4294967295
+                cdec cfg 'cdec.ini'
+                   input 'work/shard.0.0.in'
+                    refs 'work/shard.0.0.refs'
+                  output 'work/weights.0.0'
+(a dot represents 10 inputs)
+Iteration #1 of 1.
+  5
+WEIGHTS
+              Glue = +0.2663
+       WordPenalty = -0.0079042
+     LanguageModel = +0.44782
+ LanguageModel_OOV = -0.0401
+     PhraseModel_0 = -0.193
+     PhraseModel_1 = +0.71321
+     PhraseModel_2 = +0.85196
+     PhraseModel_3 = -0.43986
+     PhraseModel_4 = -0.44803
+     PhraseModel_5 = -0.0538
+     PhraseModel_6 = -0.1788
+       PassThrough = -0.1477
+        ---
+       1best avg score: 0.17521 (+0.17521)
+ 1best avg model score: 21.556 (+21.556)
+           avg # pairs: 1671.2
+        avg # rank err: 1118.6
+     avg # margin viol: 552.6
+    non0 feature count: 12
+           avg list sz: 100
+           avg f count: 11.32
+(time 0.37 min, 4.4 s/S)
+
+Writing weights file to 'work/weights.0.0' ...
+done
+
+---
+Best iteration: 1 [SCORE 'stupid_bleu'=0.17521].
+This took 0.36667 min.
diff --git a/training/dtrain/examples/parallelized/work/out.0.1 b/training/dtrain/examples/parallelized/work/out.0.1
new file mode 100644
index 00000000..e2bd6649
--- /dev/null
+++ b/training/dtrain/examples/parallelized/work/out.0.1
@@ -0,0 +1,62 @@
+                cdec cfg 'cdec.ini'
+Loading the LM will be faster if you build a binary file.
+Reading ../example/nc-wmt11.en.srilm.gz
+----5---10---15---20---25---30---35---40---45---50---55---60---65---70---75---80---85---90---95--100
+****************************************************************************************************
+Seeding random number sequence to 2767202922
+
+dtrain
+Parameters:
+                       k 100
+                       N 4
+                       T 1
+                  scorer 'stupid_bleu'
+             sample from 'kbest'
+                  filter 'uniq'
+           learning rate 0.0001
+                   gamma 0
+             loss margin 1
+                   pairs 'XYX'
+                   hi lo 0.1
+          pair threshold 0
+          select weights 'last'
+                  l1 reg 0 'none'
+               max pairs 4294967295
+                cdec cfg 'cdec.ini'
+                   input 'work/shard.0.0.in'
+                    refs 'work/shard.0.0.refs'
+                  output 'work/weights.0.1'
+              weights in 'work/weights.0'
+(a dot represents 10 inputs)
+Iteration #1 of 1.
+  5
+WEIGHTS
+              Glue = -0.2699
+       WordPenalty = +0.080605
+     LanguageModel = -0.026572
+ LanguageModel_OOV = -0.30025
+     PhraseModel_0 = -0.32076
+     PhraseModel_1 = +0.67451
+     PhraseModel_2 = +0.92
+     PhraseModel_3 = -0.36402
+     PhraseModel_4 = -0.592
+     PhraseModel_5 = -0.0269
+     PhraseModel_6 = -0.28755
+       PassThrough = -0.33285
+        ---
+       1best avg score: 0.26638 (+0.26638)
+ 1best avg model score: 53.197 (+53.197)
+           avg # pairs: 2028.6
+        avg # rank err: 998.2
+     avg # margin viol: 918.8
+    non0 feature count: 12
+           avg list sz: 100
+           avg f count: 10.496
+(time 0.32 min, 3.8 s/S)
+
+Writing weights file to 'work/weights.0.1' ...
+done
+
+---
+Best iteration: 1 [SCORE 'stupid_bleu'=0.26638].
+This took 0.31667 min.
diff --git a/training/dtrain/examples/parallelized/work/out.1.0 b/training/dtrain/examples/parallelized/work/out.1.0
new file mode 100644
index 00000000..6e790e38
--- /dev/null
+++ b/training/dtrain/examples/parallelized/work/out.1.0
@@ -0,0 +1,61 @@
+                cdec cfg 'cdec.ini'
+Loading the LM will be faster if you build a binary file.
+Reading ../example/nc-wmt11.en.srilm.gz
+----5---10---15---20---25---30---35---40---45---50---55---60---65---70---75---80---85---90---95--100
+****************************************************************************************************
+Seeding random number sequence to 1432415010
+
+dtrain
+Parameters:
+                       k 100
+                       N 4
+                       T 1
+                  scorer 'stupid_bleu'
+             sample from 'kbest'
+                  filter 'uniq'
+           learning rate 0.0001
+                   gamma 0
+             loss margin 1
+                   pairs 'XYX'
+                   hi lo 0.1
+          pair threshold 0
+          select weights 'last'
+                  l1 reg 0 'none'
+               max pairs 4294967295
+                cdec cfg 'cdec.ini'
+                   input 'work/shard.1.0.in'
+                    refs 'work/shard.1.0.refs'
+                  output 'work/weights.1.0'
+(a dot represents 10 inputs)
+Iteration #1 of 1.
+  5
+WEIGHTS
+              Glue = -0.3815
+       WordPenalty = +0.20064
+     LanguageModel = +0.95304
+ LanguageModel_OOV = -0.264
+     PhraseModel_0 = -0.22362
+     PhraseModel_1 = +0.12254
+     PhraseModel_2 = +0.26328
+     PhraseModel_3 = +0.38018
+     PhraseModel_4 = -0.48654
+     PhraseModel_5 = +0
+     PhraseModel_6 = -0.3645
+       PassThrough = -0.2216
+        ---
+       1best avg score: 0.10863 (+0.10863)
+ 1best avg model score: -4.9841 (-4.9841)
+           avg # pairs: 1345.4
+        avg # rank err: 822.4
+     avg # margin viol: 501
+    non0 feature count: 11
+           avg list sz: 100
+           avg f count: 11.814
+(time 0.45 min, 5.4 s/S)
+
+Writing weights file to 'work/weights.1.0' ...
+done
+
+---
+Best iteration: 1 [SCORE 'stupid_bleu'=0.10863].
+This took 0.45 min.
diff --git a/training/dtrain/examples/parallelized/work/out.1.1 b/training/dtrain/examples/parallelized/work/out.1.1
new file mode 100644
index 00000000..0b984761
--- /dev/null
+++ b/training/dtrain/examples/parallelized/work/out.1.1
@@ -0,0 +1,62 @@
+                cdec cfg 'cdec.ini'
+Loading the LM will be faster if you build a binary file.
+Reading ../example/nc-wmt11.en.srilm.gz
+----5---10---15---20---25---30---35---40---45---50---55---60---65---70---75---80---85---90---95--100
+****************************************************************************************************
+Seeding random number sequence to 1771918374
+
+dtrain
+Parameters:
+                       k 100
+                       N 4
+                       T 1
+                  scorer 'stupid_bleu'
+             sample from 'kbest'
+                  filter 'uniq'
+           learning rate 0.0001
+                   gamma 0
+             loss margin 1
+                   pairs 'XYX'
+                   hi lo 0.1
+          pair threshold 0
+          select weights 'last'
+                  l1 reg 0 'none'
+               max pairs 4294967295
+                cdec cfg 'cdec.ini'
+                   input 'work/shard.1.0.in'
+                    refs 'work/shard.1.0.refs'
+                  output 'work/weights.1.1'
+              weights in 'work/weights.0'
+(a dot represents 10 inputs)
+Iteration #1 of 1.
+  5
+WEIGHTS
+              Glue = -0.3178
+       WordPenalty = +0.11092
+     LanguageModel = +0.17269
+ LanguageModel_OOV = -0.13485
+     PhraseModel_0 = -0.45371
+     PhraseModel_1 = +0.38789
+     PhraseModel_2 = +0.75311
+     PhraseModel_3 = -0.38163
+     PhraseModel_4 = -0.58817
+     PhraseModel_5 = -0.0269
+     PhraseModel_6 = -0.27315
+       PassThrough = -0.16745
+        ---
+       1best avg score: 0.13169 (+0.13169)
+ 1best avg model score: 24.226 (+24.226)
+           avg # pairs: 1951.2
+        avg # rank err: 985.4
+     avg # margin viol: 951
+    non0 feature count: 12
+           avg list sz: 100
+           avg f count: 11.224
+(time 0.42 min, 5 s/S)
+
+Writing weights file to 'work/weights.1.1' ...
+done
+
+---
+Best iteration: 1 [SCORE 'stupid_bleu'=0.13169].
+This took 0.41667 min.
diff --git a/training/dtrain/examples/parallelized/work/shard.0.0.in b/training/dtrain/examples/parallelized/work/shard.0.0.in
new file mode 100644
index 00000000..92f9c78e
--- /dev/null
+++ b/training/dtrain/examples/parallelized/work/shard.0.0.in
@@ -0,0 +1,5 @@
+<seg grammar="grammar/grammar.out.0.gz" id="0">europas nach rassen geteiltes haus</seg>
+<seg grammar="grammar/grammar.out.1.gz" id="1">ein gemeinsames merkmal aller extremen rechten in europa ist ihr rassismus und die tatsache , daß sie das einwanderungsproblem als politischen hebel benutzen .</seg>
+<seg grammar="grammar/grammar.out.2.gz" id="2">der lega nord in italien , der vlaams block in den niederlanden , die anhänger von le pens nationaler front in frankreich , sind beispiele für parteien oder bewegungen , die sich um das gemeinsame thema : ablehnung der zuwanderung gebildet haben und um forderung nach einer vereinfachten politik , um sie zu regeln .</seg>
+<seg grammar="grammar/grammar.out.3.gz" id="3">während individuen wie jörg haidar und jean @-@ marie le pen kommen und ( leider nicht zu bald ) wieder gehen mögen , wird die rassenfrage aus der europäischer politik nicht so bald verschwinden .</seg>
+<seg grammar="grammar/grammar.out.4.gz" id="4">eine alternde einheimische bevölkerung und immer offenere grenzen vermehren die rassistische zersplitterung in den europäischen ländern .</seg>
diff --git a/training/dtrain/examples/parallelized/work/shard.0.0.refs b/training/dtrain/examples/parallelized/work/shard.0.0.refs
new file mode 100644
index 00000000..bef68fee
--- /dev/null
+++ b/training/dtrain/examples/parallelized/work/shard.0.0.refs
@@ -0,0 +1,5 @@
+europe 's divided racial house
+a common feature of europe 's extreme right is its racism and use of the immigration issue as a political wedge .
+the lega nord in italy , the vlaams blok in the netherlands , the supporters of le pen 's national front in france , are all examples of parties or movements formed on the common theme of aversion to immigrants and promotion of simplistic policies to control them .
+while individuals like jorg haidar and jean @-@ marie le pen may come and ( never to soon ) go , the race question will not disappear from european politics anytime soon .
+an aging population at home and ever more open borders imply increasing racial fragmentation in european countries .
diff --git a/training/dtrain/examples/parallelized/work/shard.1.0.in b/training/dtrain/examples/parallelized/work/shard.1.0.in
new file mode 100644
index 00000000..b7695ce7
--- /dev/null
+++ b/training/dtrain/examples/parallelized/work/shard.1.0.in
@@ -0,0 +1,5 @@
+<seg grammar="grammar/grammar.out.5.gz" id="5">die großen parteien der rechten und der linken mitte haben sich dem problem gestellt , in dem sie den kopf in den sand gesteckt und allen aussichten zuwider gehofft haben , es möge bald verschwinden .</seg>
+<seg grammar="grammar/grammar.out.6.gz" id="6">das aber wird es nicht , wie die geschichte des rassismus in amerika deutlich zeigt .</seg>
+<seg grammar="grammar/grammar.out.7.gz" id="7">die beziehungen zwischen den rassen standen in den usa über jahrzehnte - und tun das noch heute - im zentrum der politischen debatte . das ging so weit , daß rassentrennung genauso wichtig wie das einkommen wurde , - wenn nicht sogar noch wichtiger - um politische zuneigungen und einstellungen zu bestimmen .</seg>
+<seg grammar="grammar/grammar.out.8.gz" id="8">der erste schritt , um mit der rassenfrage umzugehen ist , ursache und folgen rassistischer feindseligkeiten zu verstehen , auch dann , wenn das bedeutet , unangenehme tatsachen aufzudecken .</seg>
+<seg grammar="grammar/grammar.out.9.gz" id="9">genau das haben in den usa eine große anzahl an forschungsvorhaben in wirtschaft , soziologie , psychologie und politikwissenschaft geleistet . diese forschungen zeigten , daß menschen unterschiedlicher rasse einander deutlich weniger vertrauen .</seg>
diff --git a/training/dtrain/examples/parallelized/work/shard.1.0.refs b/training/dtrain/examples/parallelized/work/shard.1.0.refs
new file mode 100644
index 00000000..6076f6d5
--- /dev/null
+++ b/training/dtrain/examples/parallelized/work/shard.1.0.refs
@@ -0,0 +1,5 @@
+mainstream parties of the center left and center right have confronted this prospect by hiding their heads in the ground , hoping against hope that the problem will disappear .
+it will not , as america 's racial history clearly shows .
+race relations in the us have been for decades - and remain - at the center of political debate , to the point that racial cleavages are as important as income , if not more , as determinants of political preferences and attitudes .
+the first step to address racial politics is to understand the origin and consequences of racial animosity , even if it means uncovering unpleasant truths .
+this is precisely what a large amount of research in economics , sociology , psychology and political science has done for the us .
diff --git a/training/dtrain/examples/parallelized/work/weights.0 b/training/dtrain/examples/parallelized/work/weights.0
new file mode 100644
index 00000000..ddd595a8
--- /dev/null
+++ b/training/dtrain/examples/parallelized/work/weights.0
@@ -0,0 +1,12 @@
+LanguageModel	0.7004298992212881
+PhraseModel_2	0.5576194336478857
+PhraseModel_1	0.41787318415343155
+PhraseModel_4	-0.46728502545635164
+PhraseModel_3	-0.029839521598455515
+Glue	-0.05760000000000068
+PhraseModel_6	-0.2716499999999978
+PhraseModel_0	-0.20831031065605327
+LanguageModel_OOV	-0.15205000000000077
+PassThrough	-0.1846500000000006
+WordPenalty	0.09636994553433414
+PhraseModel_5	-0.026900000000000257
diff --git a/training/dtrain/examples/parallelized/work/weights.0.0 b/training/dtrain/examples/parallelized/work/weights.0.0
new file mode 100644
index 00000000..c9370b18
--- /dev/null
+++ b/training/dtrain/examples/parallelized/work/weights.0.0
@@ -0,0 +1,12 @@
+WordPenalty	-0.0079041595706392243
+LanguageModel	0.44781580828279532
+LanguageModel_OOV	-0.04010000000000042
+Glue	0.26629999999999948
+PhraseModel_0	-0.19299677809125185
+PhraseModel_1	0.71321026861732773
+PhraseModel_2	0.85195540993310537
+PhraseModel_3	-0.43986310822842656
+PhraseModel_4	-0.44802855630415955
+PhraseModel_5	-0.053800000000000514
+PhraseModel_6	-0.17879999999999835
+PassThrough	-0.14770000000000036
diff --git a/training/dtrain/examples/parallelized/work/weights.0.1 b/training/dtrain/examples/parallelized/work/weights.0.1
new file mode 100644
index 00000000..8fad3de8
--- /dev/null
+++ b/training/dtrain/examples/parallelized/work/weights.0.1
@@ -0,0 +1,12 @@
+WordPenalty	0.080605055841244472
+LanguageModel	-0.026571720531022844
+LanguageModel_OOV	-0.30024999999999141
+Glue	-0.26989999999999842
+PhraseModel_2	0.92000295209089566
+PhraseModel_1	0.67450748692470841
+PhraseModel_4	-0.5920000014976784
+PhraseModel_3	-0.36402437203127397
+PhraseModel_6	-0.28754999999999603
+PhraseModel_0	-0.32076244202907672
+PassThrough	-0.33284999999999004
+PhraseModel_5	-0.026900000000000257
diff --git a/training/dtrain/examples/parallelized/work/weights.1 b/training/dtrain/examples/parallelized/work/weights.1
new file mode 100644
index 00000000..03058a16
--- /dev/null
+++ b/training/dtrain/examples/parallelized/work/weights.1
@@ -0,0 +1,12 @@
+PhraseModel_2	0.8365578543552836
+PhraseModel_4	-0.5900840266009169
+PhraseModel_1	0.5312000609786991
+PhraseModel_0	-0.3872342271319619
+PhraseModel_3	-0.3728279676912084
+Glue	-0.2938500000000036
+PhraseModel_6	-0.2803499999999967
+PassThrough	-0.25014999999999626
+LanguageModel_OOV	-0.21754999999999702
+LanguageModel	0.07306061161169894
+WordPenalty	0.09576193325966899
+PhraseModel_5	-0.026900000000000257
diff --git a/training/dtrain/examples/parallelized/work/weights.1.0 b/training/dtrain/examples/parallelized/work/weights.1.0
new file mode 100644
index 00000000..6a6a65c1
--- /dev/null
+++ b/training/dtrain/examples/parallelized/work/weights.1.0
@@ -0,0 +1,11 @@
+WordPenalty	0.20064405063930751
+LanguageModel	0.9530439901597807
+LanguageModel_OOV	-0.26400000000000112
+Glue	-0.38150000000000084
+PhraseModel_0	-0.22362384322085468
+PhraseModel_1	0.12253609968953538
+PhraseModel_2	0.26328345736266612
+PhraseModel_3	0.38018406503151553
+PhraseModel_4	-0.48654149460854373
+PhraseModel_6	-0.36449999999999722
+PassThrough	-0.22160000000000085
diff --git a/training/dtrain/examples/parallelized/work/weights.1.1 b/training/dtrain/examples/parallelized/work/weights.1.1
new file mode 100644
index 00000000..f56ea4a2
--- /dev/null
+++ b/training/dtrain/examples/parallelized/work/weights.1.1
@@ -0,0 +1,12 @@
+WordPenalty	0.1109188106780935
+LanguageModel	0.17269294375442074
+LanguageModel_OOV	-0.13485000000000266
+Glue	-0.3178000000000088
+PhraseModel_2	0.75311275661967159
+PhraseModel_1	0.38789263503268989
+PhraseModel_4	-0.58816805170415531
+PhraseModel_3	-0.38163156335114284
+PhraseModel_6	-0.27314999999999739
+PhraseModel_0	-0.45370601223484697
+PassThrough	-0.16745000000000249
+PhraseModel_5	-0.026900000000000257
diff --git a/training/dtrain/examples/standard/README b/training/dtrain/examples/standard/README
new file mode 100644
index 00000000..ce37d31a
--- /dev/null
+++ b/training/dtrain/examples/standard/README
@@ -0,0 +1,2 @@
+Call `dtrain` from this folder with ../../dtrain -c dtrain.ini .
+
diff --git a/training/dtrain/examples/standard/cdec.ini b/training/dtrain/examples/standard/cdec.ini
new file mode 100644
index 00000000..e1edc68d
--- /dev/null
+++ b/training/dtrain/examples/standard/cdec.ini
@@ -0,0 +1,26 @@
+formalism=scfg
+add_pass_through_rules=true
+scfg_max_span_limit=15
+intersection_strategy=cube_pruning
+cubepruning_pop_limit=200
+grammar=nc-wmt11.grammar.gz
+feature_function=WordPenalty
+feature_function=KLanguageModel ./nc-wmt11.en.srilm.gz
+# all currently working feature functions for translation:
+# (with those features active that were used in the ACL paper)
+#feature_function=ArityPenalty
+#feature_function=CMR2008ReorderingFeatures
+#feature_function=Dwarf
+#feature_function=InputIndicator
+#feature_function=LexNullJump
+#feature_function=NewJump
+#feature_function=NgramFeatures
+#feature_function=NonLatinCount
+#feature_function=OutputIndicator
+feature_function=RuleIdentityFeatures
+feature_function=RuleSourceBigramFeatures
+feature_function=RuleTargetBigramFeatures
+feature_function=RuleShape
+#feature_function=SourceSpanSizeFeatures
+#feature_function=SourceWordPenalty
+#feature_function=SpanFeatures
diff --git a/training/dtrain/examples/standard/dtrain.ini b/training/dtrain/examples/standard/dtrain.ini
new file mode 100644
index 00000000..e1072d30
--- /dev/null
+++ b/training/dtrain/examples/standard/dtrain.ini
@@ -0,0 +1,24 @@
+input=./nc-wmt11.de.gz
+refs=./nc-wmt11.en.gz
+output=-                  # a weights file (add .gz for gzip compression) or STDOUT '-'
+select_weights=VOID       # output average (over epochs) weight vector
+decoder_config=./cdec.ini # config for cdec
+# weights for these features will be printed on each iteration
+print_weights=Glue WordPenalty LanguageModel LanguageModel_OOV PhraseModel_0 PhraseModel_1 PhraseModel_2 PhraseModel_3 PhraseModel_4 PhraseModel_5 PhraseModel_6 PassThrough
+# newer version of the grammar extractor use different feature names: 
+#print_weights= EgivenFCoherent SampleCountF CountEF MaxLexFgivenE MaxLexEgivenF IsSingletonF IsSingletonFE Glue WordPenalty PassThrough LanguageModel LanguageModel_OOV
+stop_after=10 # stop epoch after 10 inputs
+
+# interesting stuff
+epochs=2                # run over input 2 times
+k=100                   # use 100best lists
+N=4                     # optimize (approx) BLEU4
+scorer=stupid_bleu      # use 'stupid' BLEU+1
+learning_rate=1.0       # learning rate, don't care if gamma=0 (perceptron)
+gamma=0                 # use SVM reg
+sample_from=kbest       # use kbest lists (as opposed to forest)
+filter=uniq             # only unique entries in kbest (surface form)
+pair_sampling=XYX       #
+hi_lo=0.1               # 10 vs 80 vs 10 and 80 vs 10 here
+pair_threshold=0        # minimum distance in BLEU (here: > 0)
+loss_margin=0           # update if correctly ranked, but within this margin
diff --git a/training/dtrain/examples/standard/expected-output b/training/dtrain/examples/standard/expected-output
new file mode 100644
index 00000000..7cd09dbf
--- /dev/null
+++ b/training/dtrain/examples/standard/expected-output
@@ -0,0 +1,91 @@
+                cdec cfg './cdec.ini'
+Loading the LM will be faster if you build a binary file.
+Reading ./nc-wmt11.en.srilm.gz
+----5---10---15---20---25---30---35---40---45---50---55---60---65---70---75---80---85---90---95--100
+****************************************************************************************************
+  Example feature: Shape_S00000_T00000
+Seeding random number sequence to 2679584485
+
+dtrain
+Parameters:
+                       k 100
+                       N 4
+                       T 2
+                  scorer 'stupid_bleu'
+             sample from 'kbest'
+                  filter 'uniq'
+           learning rate 1
+                   gamma 0
+             loss margin 0
+       faster perceptron 1
+                   pairs 'XYX'
+                   hi lo 0.1
+          pair threshold 0
+          select weights 'VOID'
+                  l1 reg 0 'none'
+               max pairs 4294967295
+                cdec cfg './cdec.ini'
+                   input './nc-wmt11.de.gz'
+                    refs './nc-wmt11.en.gz'
+                  output '-'
+              stop_after 10
+(a dot represents 10 inputs)
+Iteration #1 of 2.
+ . 10
+Stopping after 10 input sentences.
+WEIGHTS
+              Glue = -576
+       WordPenalty = +417.79
+     LanguageModel = +5117.5
+ LanguageModel_OOV = -1307
+     PhraseModel_0 = -1612
+     PhraseModel_1 = -2159.6
+     PhraseModel_2 = -677.36
+     PhraseModel_3 = +2663.8
+     PhraseModel_4 = -1025.9
+     PhraseModel_5 = -8
+     PhraseModel_6 = +70
+       PassThrough = -1455
+        ---
+       1best avg score: 0.27697 (+0.27697)
+ 1best avg model score: -47918 (-47918)
+           avg # pairs: 581.9 (meaningless)
+        avg # rank err: 581.9
+     avg # margin viol: 0
+    non0 feature count: 703
+           avg list sz: 90.9
+           avg f count: 100.09
+(time 0.25 min, 1.5 s/S)
+
+Iteration #2 of 2.
+ . 10
+WEIGHTS
+              Glue = -622
+       WordPenalty = +898.56
+     LanguageModel = +8066.2
+ LanguageModel_OOV = -2590
+     PhraseModel_0 = -4335.8
+     PhraseModel_1 = -5864.4
+     PhraseModel_2 = -1729.8
+     PhraseModel_3 = +2831.9
+     PhraseModel_4 = -5384.8
+     PhraseModel_5 = +1449
+     PhraseModel_6 = +480
+       PassThrough = -2578
+        ---
+       1best avg score: 0.37119 (+0.094226)
+ 1best avg model score: -1.3174e+05 (-83822)
+           avg # pairs: 584.1 (meaningless)
+        avg # rank err: 584.1
+     avg # margin viol: 0
+    non0 feature count: 1115
+           avg list sz: 91.3
+           avg f count: 90.755
+(time 0.3 min, 1.8 s/S)
+
+Writing weights file to '-' ...
+done
+
+---
+Best iteration: 2 [SCORE 'stupid_bleu'=0.37119].
+This took 0.55 min.
diff --git a/training/dtrain/examples/standard/nc-wmt11.de.gz b/training/dtrain/examples/standard/nc-wmt11.de.gz
new file mode 100644
index 00000000..0741fd92
--- /dev/null
+++ b/training/dtrain/examples/standard/nc-wmt11.de.gz
diff --git a/training/dtrain/examples/standard/nc-wmt11.en.gz b/training/dtrain/examples/standard/nc-wmt11.en.gz
new file mode 100644
index 00000000..1c0bd401
--- /dev/null
+++ b/training/dtrain/examples/standard/nc-wmt11.en.gz
diff --git a/training/dtrain/examples/standard/nc-wmt11.en.srilm.gz b/training/dtrain/examples/standard/nc-wmt11.en.srilm.gz
new file mode 100644
index 00000000..7ce81057
--- /dev/null
+++ b/training/dtrain/examples/standard/nc-wmt11.en.srilm.gz
diff --git a/training/dtrain/examples/standard/nc-wmt11.grammar.gz b/training/dtrain/examples/standard/nc-wmt11.grammar.gz
new file mode 100644
index 00000000..ce4024a1
--- /dev/null
+++ b/training/dtrain/examples/standard/nc-wmt11.grammar.gz
diff --git a/training/dtrain/examples/toy/cdec.ini b/training/dtrain/examples/toy/cdec.ini
new file mode 100644
index 00000000..b14f4819
--- /dev/null
+++ b/training/dtrain/examples/toy/cdec.ini
@@ -0,0 +1,3 @@
+formalism=scfg
+add_pass_through_rules=true
+grammar=grammar.gz
diff --git a/training/dtrain/examples/toy/dtrain.ini b/training/dtrain/examples/toy/dtrain.ini
new file mode 100644
index 00000000..cd715f26
--- /dev/null
+++ b/training/dtrain/examples/toy/dtrain.ini
@@ -0,0 +1,13 @@
+decoder_config=cdec.ini
+input=src
+refs=tgt
+output=-
+print_weights=logp shell_rule house_rule small_rule little_rule PassThrough
+k=4
+N=4
+epochs=2
+scorer=bleu
+sample_from=kbest
+filter=uniq
+pair_sampling=all
+learning_rate=1
diff --git a/training/dtrain/examples/toy/expected-output b/training/dtrain/examples/toy/expected-output
new file mode 100644
index 00000000..1da2aadd
--- /dev/null
+++ b/training/dtrain/examples/toy/expected-output
@@ -0,0 +1,77 @@
+Warning: hi_lo only works with pair_sampling XYX.
+                cdec cfg 'cdec.ini'
+Seeding random number sequence to 1664825829
+
+dtrain
+Parameters:
+                       k 4
+                       N 4
+                       T 2
+                  scorer 'bleu'
+             sample from 'kbest'
+                  filter 'uniq'
+           learning rate 1
+                   gamma 0
+             loss margin 0
+                   pairs 'all'
+          pair threshold 0
+          select weights 'last'
+                  l1 reg 0 'none'
+               max pairs 4294967295
+                cdec cfg 'cdec.ini'
+                   input 'src'
+                    refs 'tgt'
+                  output '-'
+(a dot represents 10 inputs)
+Iteration #1 of 2.
+  2
+WEIGHTS
+              logp = +0
+        shell_rule = -1
+        house_rule = +2
+        small_rule = -2
+       little_rule = +3
+       PassThrough = -5
+        ---
+       1best avg score: 0.5 (+0.5)
+ 1best avg model score: 2.5 (+2.5)
+           avg # pairs: 4
+        avg # rank err: 1.5
+     avg # margin viol: 0
+    non0 feature count: 6
+           avg list sz: 4
+           avg f count: 2.875
+(time 0 min, 0 s/S)
+
+Iteration #2 of 2.
+  2
+WEIGHTS
+              logp = +0
+        shell_rule = -1
+        house_rule = +2
+        small_rule = -2
+       little_rule = +3
+       PassThrough = -5
+        ---
+       1best avg score: 1 (+0.5)
+ 1best avg model score: 5 (+2.5)
+           avg # pairs: 5
+        avg # rank err: 0
+     avg # margin viol: 0
+    non0 feature count: 6
+           avg list sz: 4
+           avg f count: 3
+(time 0 min, 0 s/S)
+
+Writing weights file to '-' ...
+house_rule	2
+little_rule	3
+Glue	-4
+PassThrough	-5
+small_rule	-2
+shell_rule	-1
+done
+
+---
+Best iteration: 2 [SCORE 'bleu'=1].
+This took 0 min.
diff --git a/training/dtrain/examples/toy/grammar.gz b/training/dtrain/examples/toy/grammar.gz
new file mode 100644
index 00000000..8eb0d29e
--- /dev/null
+++ b/training/dtrain/examples/toy/grammar.gz
diff --git a/training/dtrain/examples/toy/src b/training/dtrain/examples/toy/src
new file mode 100644
index 00000000..87e39ef2
--- /dev/null
+++ b/training/dtrain/examples/toy/src
@@ -0,0 +1,2 @@
+ich sah ein kleines haus
+ich fand ein kleines haus
diff --git a/training/dtrain/examples/toy/tgt b/training/dtrain/examples/toy/tgt
new file mode 100644
index 00000000..174926b3
--- /dev/null
+++ b/training/dtrain/examples/toy/tgt
@@ -0,0 +1,2 @@
+i saw a little house
+i found a little house