summaryrefslogtreecommitdiff
path: root/dtrain/test/example
diff options
context:
space:
mode:
authorChris Dyer <redpony@gmail.com>2012-11-05 10:25:54 -0800
committerChris Dyer <redpony@gmail.com>2012-11-05 10:25:54 -0800
commit2eb77b77d28f6d9dc88ecb6ca999994bdb555445 (patch)
tree1e19af403041bd86933ce88fa1a70030ab100ca2 /dtrain/test/example
parentc615c37501fa8576584a510a9d2bfe2fdd5bace7 (diff)
parent0c54220adfaada6ad1e2d54f31a9895da35127fd (diff)
Merge pull request #10 from pks/master
renamed RuleNgramFeatures
Diffstat (limited to 'dtrain/test/example')
-rw-r--r--dtrain/test/example/cdec.ini3
-rw-r--r--dtrain/test/example/dtrain.ini6
-rw-r--r--dtrain/test/example/expected-output130
3 files changed, 52 insertions, 87 deletions
diff --git a/dtrain/test/example/cdec.ini b/dtrain/test/example/cdec.ini
index 6642107f..d5955f0e 100644
--- a/dtrain/test/example/cdec.ini
+++ b/dtrain/test/example/cdec.ini
@@ -17,7 +17,8 @@ feature_function=KLanguageModel test/example/nc-wmt11.en.srilm.gz
#feature_function=NonLatinCount
#feature_function=OutputIndicator
feature_function=RuleIdentityFeatures
-feature_function=RuleNgramFeatures
+feature_function=RuleSourceBigramFeatures
+feature_function=RuleTargetBigramFeatures
feature_function=RuleShape
#feature_function=SourceSpanSizeFeatures
#feature_function=SourceWordPenalty
diff --git a/dtrain/test/example/dtrain.ini b/dtrain/test/example/dtrain.ini
index c8ac7c3f..72d50ca1 100644
--- a/dtrain/test/example/dtrain.ini
+++ b/dtrain/test/example/dtrain.ini
@@ -1,6 +1,6 @@
input=test/example/nc-wmt11.1k.gz # use '-' for STDIN
output=- # a weights file (add .gz for gzip compression) or STDOUT '-'
-select_weights=VOID # don't output weights
+select_weights=VOID # don't output weights
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
@@ -8,11 +8,11 @@ tmp=/tmp
stop_after=10 # stop epoch after 10 inputs
# interesting stuff
-epochs=3 # run over input 3 times
+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=0.0001 # learning rate
+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)
diff --git a/dtrain/test/example/expected-output b/dtrain/test/example/expected-output
index 25d2c069..05326763 100644
--- a/dtrain/test/example/expected-output
+++ b/dtrain/test/example/expected-output
@@ -1,31 +1,20 @@
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
+Seeding random number sequence to 2912000813
dtrain
Parameters:
k 100
N 4
- T 3
+ T 2
scorer 'stupid_bleu'
sample from 'kbest'
filter 'uniq'
- learning rate 0.0001
+ learning rate 1
gamma 0
loss margin 0
pairs 'XYX'
@@ -33,93 +22,68 @@ Parameters:
pair threshold 0
select weights 'VOID'
l1 reg 0 'none'
+ max pairs 4294967295
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.
+Iteration #1 of 2.
. 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
+ Glue = -637
+ WordPenalty = +1064
+ LanguageModel = +1175.3
+ LanguageModel_OOV = -1437
+ PhraseModel_0 = +1935.6
+ PhraseModel_1 = +2499.3
+ PhraseModel_2 = +964.96
+ PhraseModel_3 = +1410.8
+ PhraseModel_4 = -5977.9
+ PhraseModel_5 = +522
+ PhraseModel_6 = +1089
+ PassThrough = -1308
---
- 1best avg score: 0.16872 (+0.16872)
- 1best avg model score: -1.8276 (-1.8276)
- avg # pairs: 1121.1
- avg # rank err: 555.6
+ 1best avg score: 0.16963 (+0.16963)
+ 1best avg model score: 64485 (+64485)
+ avg # pairs: 1494.4
+ avg # rank err: 702.6
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
+ non0 feature count: 528
+ avg list sz: 85.7
+ avg f count: 102.75
(time 0.083 min, 0.5 s/S)
-Iteration #3 of 3.
+Iteration #2 of 2.
. 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
+ Glue = -1196
+ WordPenalty = +809.52
+ LanguageModel = +3112.1
+ LanguageModel_OOV = -1464
+ PhraseModel_0 = +3895.5
+ PhraseModel_1 = +4683.4
+ PhraseModel_2 = +1092.8
+ PhraseModel_3 = +1079.6
+ PhraseModel_4 = -6827.7
+ PhraseModel_5 = -888
+ PhraseModel_6 = +142
+ PassThrough = -1335
---
- 1best avg score: 0.2962 (+0.11225)
- 1best avg model score: -36.274 (-39.479)
- avg # pairs: 1109.6
- avg # rank err: 515.9
+ 1best avg score: 0.277 (+0.10736)
+ 1best avg model score: -3110.5 (-67595)
+ avg # pairs: 1144.2
+ avg # rank err: 529.1
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)
+ non0 feature count: 859
+ avg list sz: 74.9
+ avg f count: 112.84
+(time 0.067 min, 0.4 s/S)
Writing weights file to '-' ...
done
---
-Best iteration: 3 [SCORE 'stupid_bleu'=0.2962].
-This took 0.26667 min.
+Best iteration: 2 [SCORE 'stupid_bleu'=0.277].
+This took 0.15 min.