Merge branch 'master' of https://github.com/pks/cdec-dtrain

3 files changed, 129 insertions, 3 deletions

diff --git a/dtrain/test/example/README b/dtrain/test/example/README
index b3ea5f06..6937b11b 100644
--- a/dtrain/test/example/README
+++ b/dtrain/test/example/README
@@ -1,8 +1,8 @@
 Small example of input format for distributed training.
 Call dtrain from cdec/dtrain/ with ./dtrain -c test/example/dtrain.ini .
 
-For this to work, disable '#define DTRAIN_LOCAL' from dtrain.h
+For this to work, undef 'DTRAIN_LOCAL' in dtrain.h
 and recompile.
 
-Data is here: http://simianer.de/dtrain
+Data is here: http://simianer.de/#dtrain
 
diff --git a/dtrain/test/example/dtrain.ini b/dtrain/test/example/dtrain.ini
index f87ee9cf..c8ac7c3f 100644
--- a/dtrain/test/example/dtrain.ini
+++ b/dtrain/test/example/dtrain.ini
@@ -5,7 +5,7 @@ decoder_config=test/example/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
 tmp=/tmp
-stop_after=10 # stop epoch after 20 inputs
+stop_after=10 # stop epoch after 10 inputs
 
 # interesting stuff
 epochs=3                # run over input 3 times
@@ -19,3 +19,4 @@ 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 (this will still only use pairs with diff > 0)
+loss_margin=0
diff --git a/dtrain/test/example/expected-output b/dtrain/test/example/expected-output
new file mode 100644
index 00000000..25d2c069
--- /dev/null
+++ b/dtrain/test/example/expected-output
@@ -0,0 +1,125 @@
+                cdec cfg 'test/example/cdec.ini'
+feature: WordPenalty (no config parameters)
+State is 0 bytes for feature WordPenalty
+feature: KLanguageModel (with config parameters 'test/example/nc-wmt11.en.srilm.gz')
+Loading the LM will be faster if you build a binary file.
+Reading test/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
+****************************************************************************************************
+Loaded 5-gram KLM from test/example/nc-wmt11.en.srilm.gz (MapSize=49581)
+State is 98 bytes for feature KLanguageModel test/example/nc-wmt11.en.srilm.gz
+feature: RuleIdentityFeatures (no config parameters)
+State is 0 bytes for feature RuleIdentityFeatures
+feature: RuleNgramFeatures (no config parameters)
+State is 0 bytes for feature RuleNgramFeatures
+feature: RuleShape (no config parameters)
+  Example feature: Shape_S00000_T00000
+State is 0 bytes for feature RuleShape
+Seeding random number sequence to 1072059181
+
+dtrain
+Parameters:
+                       k 100
+                       N 4
+                       T 3
+                 scorer 'stupid_bleu'
+             sample from 'kbest'
+                  filter 'uniq'
+           learning rate 0.0001
+                   gamma 0
+             loss margin 0
+                   pairs 'XYX'
+                   hi lo 0.1
+          pair threshold 0
+          select weights 'VOID'
+                  l1 reg 0 'none'
+                cdec cfg 'test/example/cdec.ini'
+                   input 'test/example/nc-wmt11.1k.gz'
+                  output '-'
+              stop_after 10
+(a dot represents 10 inputs)
+Iteration #1 of 3.
+ . 10
+Stopping after 10 input sentences.
+WEIGHTS
+              Glue = -0.0293
+       WordPenalty = +0.049075
+     LanguageModel = +0.24345
+ LanguageModel_OOV = -0.2029
+     PhraseModel_0 = +0.0084102
+     PhraseModel_1 = +0.021729
+     PhraseModel_2 = +0.014922
+     PhraseModel_3 = +0.104
+     PhraseModel_4 = -0.14308
+     PhraseModel_5 = +0.0247
+     PhraseModel_6 = -0.012
+       PassThrough = -0.2161
+        ---
+       1best avg score: 0.16872 (+0.16872)
+ 1best avg model score: -1.8276 (-1.8276)
+           avg # pairs: 1121.1
+        avg # rank err: 555.6
+     avg # margin viol: 0
+    non0 feature count: 277
+           avg list sz: 77.2
+           avg f count: 90.96
+(time 0.1 min, 0.6 s/S)
+
+Iteration #2 of 3.
+ . 10
+WEIGHTS
+              Glue = -0.3526
+       WordPenalty = +0.067576
+     LanguageModel = +1.155
+ LanguageModel_OOV = -0.2728
+     PhraseModel_0 = -0.025529
+     PhraseModel_1 = +0.095869
+     PhraseModel_2 = +0.094567
+     PhraseModel_3 = +0.12482
+     PhraseModel_4 = -0.36533
+     PhraseModel_5 = +0.1068
+     PhraseModel_6 = -0.1517
+       PassThrough = -0.286
+        ---
+       1best avg score: 0.18394 (+0.015221)
+ 1best avg model score: 3.205 (+5.0326)
+           avg # pairs: 1168.3
+        avg # rank err: 594.8
+     avg # margin viol: 0
+    non0 feature count: 543
+           avg list sz: 77.5
+           avg f count: 85.916
+(time 0.083 min, 0.5 s/S)
+
+Iteration #3 of 3.
+ . 10
+WEIGHTS
+              Glue = -0.392
+       WordPenalty = +0.071963
+     LanguageModel = +0.81266
+ LanguageModel_OOV = -0.4177
+     PhraseModel_0 = -0.2649
+     PhraseModel_1 = -0.17931
+     PhraseModel_2 = +0.038261
+     PhraseModel_3 = +0.20261
+     PhraseModel_4 = -0.42621
+     PhraseModel_5 = +0.3198
+     PhraseModel_6 = -0.1437
+       PassThrough = -0.4309
+        ---
+       1best avg score: 0.2962 (+0.11225)
+ 1best avg model score: -36.274 (-39.479)
+           avg # pairs: 1109.6
+        avg # rank err: 515.9
+     avg # margin viol: 0
+    non0 feature count: 741
+           avg list sz: 77
+           avg f count: 88.982
+(time 0.083 min, 0.5 s/S)
+
+Writing weights file to '-' ...
+done
+
+---
+Best iteration: 3 [SCORE 'stupid_bleu'=0.2962].
+This took 0.26667 min.