Development in progress...
Given train.txt
, dev.m2
, and test.m2
:
- Train model
- Extract features from
train.txt
producingtrain.feats
andtrain.cwords
- Train VW classifier on
train.feats
creating modelmodel.vw
- Extract features from
- Tune threshold parameter
- Create file
dev.txt
with parallel sentences fromdev.m2
- Extract features from
dev.txt
producingdev.feats
anddev.cwords
- Run VW classifier on
dev.feats
and modelmodel.vw
producingdev.pred
- Perform grid search to find best threshold parameter
- Apply predictions
dev.thr.pred
intodev.thr.out
usingdev.cwords
- Run M2 scorer on
dev.thr.out
- Apply predictions
- Create file
- Evaluate
- Create file
test.txt
with parallel sentences fromtest.m2
- Extract features from
test.txt
producingtest.feats
andtest.cwords
- Run VW classifier on
test.feats
and modelmodel.vw
producingtest.pred
- Apply predictions
test.pred
intotest.out
usingtest.cwords
- Run M2 scorer on
test.out
- Create file