sungjinl / ns-dial

An Interpretable Neuro-Symbolic Framework for Task-Oriented Dialogue Generation

NS-Dial

This is the code repository for the paper: An Interpretable Neuro-Symbolic Framework for Task-Oriented Dialogue Generation. ACL 2022.

Framework

Dependencies

• Pytorch 1.0.0
• cudatoolkit 10.0.130
• cudnn 7.6.5
• tqdm 4.54.1
• numpy 1.19.2
• python 3.6.10

Training

We created train.py to train the models. For SMD dataset, you can run:

python train.py -ds=kvr -bsz=8 -hdd=128 -lr=0.001 -dr=0.2 -evalp=10 -max_neg_cnt=5 -max_depth=3

For MultiWOZ 2.1 dataset, you can run:

python train.py -ds=multiwoz -bsz=8 -hdd=128 -lr=0.001 -dr=0.2 -evalp=10 -max_neg_cnt=5 -max_depth=3

While training, the model with the best validation results is stored. If you want to reuse a model, please add -path=path_name_model to the call. The model is evaluated by BLEU and Entity F1.

Testing

We created test.py to restore the checkpoints and test the models. For SMD dataset, you can run:

python test.py -path=<path_to_saved_model> -ds=kvr -lr=0.001 -dr=0.2 -max_depth=3

For MultiWOZ 2.1 dataset, you can run:

python test.py -path=<path_to_saved_model> -ds=multiwoz -lr=0.001 -dr=0.2 -max_depth=3