TimeTuner is a general visual analytics framework. which is designed to help analysts understand how model behaviors are associated with localized correlations, stationarity, and granularity of time-series representations. In our work, we instantiate TimeTuner with two transformation methods of smoothing and sampling, and demonstrate its applicability on real-world time-series forecasting of univariate sunspots and multivariate air pollutants.
- Front-end
- Vue.js
- Element Plus
- Backend
- Flask
- Python
- Frontend:
- Vue.js 3
- npm
- Backend:
- Python: version 3.8 or higher
- SHAP: version 0.41.0
- pandas: version 1.5.3
- tensorflow: version 2.10.0
- keras: version 2.10.0
- scikit-learn: version 1.2.2
- numpy: version 1.24.3
- Flask: version 2.2.0
- Enter the folder
cd <your-project-path>/TimeTunerSystem/Frontend
- Install NPM packages
npm install
- Run the Frontend
npm run dev
- Enter the folder
cd <your-project-path>/TimeTunerSystem/Backend
- Run the Backend
set FLASK_APP=app.py flask run // or python app.py
@article{hao2023timetuner,
title={TimeTuner: Diagnosing Time Representations for Time-Series Forecasting with Counterfactual Explanations},
author={Hao, Jianing and Shi, Qing and Ye, Yilin and Zeng, Wei},
journal={arXiv preprint arXiv:2307.09916},
year={2023}
}