Codebase for the NAACL HLT 2018 paper: Sluice Resolution without Hand-Crafted Features over Brittle Syntax trees.
To train and evaluate:
> python3 mains/<model>_train_eval.py
model: rhs, sg or base. F.ex:
> python3 mains/rhs_train_eval.py
To evaluate on dialogue:
> python3 mains/eval_dialogue -l <log> -m <model>
<model> must be trained before it can be evaluated on dialogue. F.ex:
> python3 mains/eval_dialogue -l tensorboard/20180415_230245 -m base
embeddings.npy and ids.npy in the data/glove directory must be unzip before training.
Run the command below to install depedencies.
> pip install -r requirements.txt
The dependencies are:
- absl-py==0.1.13
- astor==0.6.2
- bleach==1.5.0
- gast==0.2.0
- grpcio==1.10.1
- html5lib==0.9999999
- Markdown==2.6.11
- numpy==1.14.2
- protobuf==3.5.2.post1
- six==1.11.0
- tensorboard==1.7.0
- tensorflow==1.7.0 # Note requirement.txt is not gpu supported version
- termcolor==1.1.0
- Werkzeug==0.14.1
The structure of the repo is:
├── configs
├── data
│ ├── auxiliaries
│ ├── esc
│ ├── opensub
│ └── glove
├── load
│ ├── core
│ └── readers
├── mains
├── metrics
└── nns
└── custom_tf