This repository contains the code for 'Opponent Modeling in Negotiation Dialogues by Related Data Adaptation'.
Our work has been accepted at Findings of NAACL 2022.
Please direct all queries to the first author: Kushal Chawla (kchawla@usc.edu).
- All experiments reported in the paper were performed using Python 3.9.7 on a single V100 GPU (not tested for multi-gpu training). In addition, we used:
pytorch 1.9.1
cudatoolkit 11.1.74
pytorch-lightning 1.4.8
transformers 4.11.0
- This code has also been tested on a single A100 GPU with the available latest installations based on the commands below. This testing used Python 3.9.12, Pytorch 1.11.0, Pytorch Lightning 1.6.2, and transformers 4.18.0.
In order to quickly set up a working environment, follow the steps below:
# Install anaconda
conda create -n oppmodeling anaconda
# Activate environment
conda activate oppmodeling
# Install PyTorch
conda install pytorch torchvision torchaudio cudatoolkit=11.3 -c pytorch -c conda-forge
# Install other dependencies using pip (inside the Anaconda environment)
pip install pytorch-lightning
pip install transformers
pip install datasets
pip install -U scikit-learn
# When inside the root folder containing setup.py
pip install -e .
Create an environment variable $OppModelingStorage and point it to the storage directory in this root folder.
#Add the following to ~/.bashrc file
export OppModelingStorage="/path/to/storage"
# Refresh/restart the terminal
source ~/.bashrc
Download base versions of BERT, RoBERTa, and GPT2 pretrained models from Huggingface and store them under storage/misc/pretrained/*_pretrained folders. Finally, files such as pytorch_model.bin, config.json, vocab.json and so on will directly come under *_pretrained/ folders.
This can be easily done from a Python shell using AutoModel.from_pretrained and save_pretrained functionality from Huggingface.
Tokenizers have already been provided in this repo in the .pt format under storage/misc/pretrained/.
Use the provided training script as an example (after activating the conda environment):
chmod +x training.sh
./training.sh
This script stores the model inside storage/logs/hierarchical/ROOT_EXPT_DIR/hierarchical/version_0/.
Use the provided evaluation script as an example (after activating the conda environment):
chmod +x evaluation.sh
./evaluation.sh
The evaluation script creates a complete profile with a number of metrics at storage/logs/hierarchical/ROOT_EXPT_DIR/hierarchical/version_0/evaluation_analysis/.
@inproceedings{chawla2022opponent,
title={Opponent Modeling in Negotiation Dialogues by Related Data Adaptation},
author={Chawla, Kushal and Lucas, Gale and May, Jonathan and Gratch, Jonathan},
booktitle={Findings of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies},
year={2022}
}
Please refer to the LICENSE file in the root directory for more details.