This is the official repository for SocioDojo.
- First clone the directory.
git submodule init; git submodule update
(If showing error of no permission, need to first add a new SSH key to your GitHub account.)
- Install dependencies.
Create a new environment using conda, with Python >= 3.10.6 Install PyTorch (version >= 2.0.0). The repo is tested with PyTorch version of 1.10.1 and there is no guarentee that other version works. Then install other dependencies via:
pip install -r requirements.txt
- Download dataset.
Download the dataset here and unzip in the Env folder.
Here we detail the repo's structure:
- Agent: code for Analyst-Assistant-Actuator architecture and Hypothesis & Proof prompting
- Env: code for SocioDojo Corpus: Download the data above and unzip in this folder, it should has a structure like:
- TS: time series
- IS: information source
- KB: knowledge base
- run.py: the script for running experiments
- config.py: you may edit configurations here
A demo is available here. The demo used minute-level financial data only.
If you find our work and/or our code useful, please cite us via:
@inproceedings{
cheng2024sociodojo,
title={SocioDojo: Building Lifelong Analytical Agents with Real-world Text and Time Series},
author={Junyan Cheng and Peter Chin},
booktitle={The Twelfth International Conference on Learning Representations},
year={2024},
url={https://openreview.net/forum?id=s9z0HzWJJp}
}