search

percyliang / sempre

Semantic Parser with Execution
Other
828 stars 301 forks source link

How to run overnight samples interactively? #187

Open phkorn opened 5 years ago

phkorn commented 5 years ago

I am a bit confused on how to use the trained model (e.g. overnight/basketball) interactively.

Is the following command correct? ./run @mode=simple -languageAnalyzer corenlp.CoreNLPAnalyzer -Grammar.inPaths overnight/basketball.grammar -Builder.inParamsPath state/execs/20.exec/params -Builder.parser FloatingParser

If i try the utterance: what players made less than three assists over a season

I get 0 candidates

I would be nice to provide some example ./run calls for overnight

brinaseidel commented 5 years ago

I also had this question - did you solve it? Can anyone provide an example for how to use the overnight trained models interactively?

ppasupat commented 5 years ago

I'm not familiar with overnight, but from their CodaLab sheet you might need to use the overnight mode instead of simple:

./run @mode=overnight @domain=basketball \
  -Learner.maxTrainIters 0 -dataset.inPaths "" -interactive \
  -Builder.inParamsPath state/execs/20.exec/params

The overnight mode should load the correct parser and grammar.

phkorn commented 5 years ago

I also had this question - did you solve it? Can anyone provide an example for how to use the overnight trained models interactively?

I never got it really working and finally moved on

brinaseidel commented 5 years ago

Thanks for the suggestion. The code you provided is throwing some errors but for me but it's probably not worth anyone's time to debug it - I also figured out a workaround for my purposes (building an interactive parser on a new domain) by modifying @mode=simple.