This is code for the paper "TA&AT: Enhancing Task-Oriented Dialog with Turn-Level Auxiliary Tasks and Action-Tree Based Scheduled Sampling".
First download the preprocessed datasets and TreeBase for each MultiWOZ dataset from our provided google drive links. (https://drive.google.com/file/d/1wBbZKYpweCKbwVHIJLV3DxcJh-KX_UXc/view?usp=sharing), unzip it in this directory.
Our python version is 3.8.8.
The package can be installed by running the following command.
pip install -r requirements.txt
python -m spacy download en_core_web_sm
Our implementation supports a single GPU. Please use smaller batch size and more accumulation steps to ensure the toal batch size if out-of-memory error raises.
sh scripts/train.sh
sh scripts/test.sh
All checkpoints are saved in $MODEL_DIR with names such as 'ckpt-epoch10'.
The result file ($MODEL_OUTPUT) will be saved in the checkpoint directory.
For the MultiWOZ benchmark, we recommend to use standardized evaluation script.
# convert format for the the standardized evaluation, before running it, you should first change the variable 'file_path' to your raw inference file path, and change the variable 'file_path_1' to your target output file path.
python convert_to_standard.py
# clone the standardized evaluation repository
git clone https://github.com/Tomiinek/MultiWOZ_Evaluation
cd MultiWOZ_Evaluation
pip install -r requirements.txt
# do standardized evaluation
python evaluate.py -i $CONVERTED_MODEL_OUTPUT -b -s -r
This code is based on the released code (https://github.com/bepoetree/MTTOD) for "Improving End-to-End Task-Oriented Dialogue System with A Simple Auxiliary Task" which distributed under Apache License Version 2.0. Copyright 2021- Yohan Lee.
For the pre-trained language model, we use huggingface's Transformer (https://huggingface.co/transformers/index.html#), which distributed under Apache License Version 2.0. Copyright 2018- The Hugging Face team. All rights reserved.