About author

Hi, I am Xinyu Chen (陈新宇), a Ph.D. candidate at Polytechnique Montreal affiliated with University of Montreal in Canada. Currently, I am leading an innovative and interesting GitHub project with Prof. Lijun Sun and Prof. Nicolas Saunier. That is transdim (GitHub repository: https://github.com/xinychen/transdim), which is a problem-oriented project for transportation data imputation and prediction. This open-source project focuses on (1) handling missing data problems with various missing patterns in the spatiotemporal settings, (2) performing time series forecasting on large-scale, high-dimensional, and multidimensional data, and (3) performing time series forecasting in the presence of missing values. The goals of this open-source project include (1) providing some well-defined data modeling problems, (2) building a platform for gathering some open-source data sets, and (3) providing some Python implementation for machine learning models. Until now, it has covered the Python implementation of a manifold of machine learning models. In the meanwhile, my collaborators have developed some tensor learning models with me for spatiotemporal data imputation and forecasting.

News

• 2022/7: transdim gets 800+ stars and 240+ forks on GitHub.
• 2021/12: transdim gets 700+ stars and 220+ forks on GitHub.
• 2021/07: Our paper “Low-rank autoregressive tensor completion for spatiotemporal traffic data imputation” (authors: Xinyu Chen, Mengying Lei, Nicolas Saunier, Lijun Sun) is accepted to IEEE Transactions on Intelligent Transportation Systems. The PDF is available at arXiv, and the Python implementation is available at transdim/imputer.
• 2021/07: Our paper “Low-rank autoregressive tensor completion for spatiotemporal traffic data imputation” (authors: Xinyu Chen, Mengying Lei, Nicolas Saunier, Lijun Sun) is accepted to The 7th SIGKDD Workshop on Mining and Learning from Time Series (MiLeTS) at KDD 2021 for an oral presentation.
• 2021/06: transdim gets 600+ stars and 190+ forks on GitHub.
• 2021/05: Our paper “Scalable low-rank tensor learning for spatiotemporal traffic data imputation” (authors: Xinyu Chen, Yixian Chen, Nicolas Saunier, Lijun Sun) is accepted to Transportation Research Part C: Emerging Technologies. The PDF is available at arXiv (or DOI), and the Python implementation is available at transdim/large-imputer.
• 2021/04: Our presentation slides for the paper “Bayesian temporal factorization for multidimensional time series prediction” are now publicly available at Zenodo. These slides were presented at the course IFT 6760A: Matrix and tensor factorization techniques for machine learning (UdeM & Mila lab) in March 18, 2021.
• 2021/03: Our paper “Bayesian temporal factorization for multidimensional time series prediction” (authors: Xinyu Chen, Lijun Sun) is accepted to IEEE Transactions on Pattern Analysis and Machine Intelligence. The PDF is available at arXiv (or see DOI), and the Python implementation is available at transdim/imputer.
• 2021/02: transdim gets 500+ stars and 150+ forks on GitHub.
• 2020/05: transdim gets 300+ stars and 100+ forks on GitHub.
• 2020/05: Our paper “A nonconvex low-rank tensor completion for spatiotemporal traffic data imputation” (authors: Xinyu Chen, Jinming Yang, Lijun Sun) is accepted to Transportation Research Part C: Emerging Technologies. The PDF is available at arXiv (or see DOI), and the Python implementation is available at Imputation-LRTC-TNN.ipynb.
• 2020/03: transdim project is awarded by IVADO Excellence PhD funding (proposal: “City-scale traffic data imputation and forecasting with tensor learning”).
• 2018/09: Xinyu Chen starts transdim project with personal interest at Sun Yat-Sen University. He thinks transdim project would help a lot for the research community.

Highlights

• Several publicly available spatiotemporal data sets, and most of them are collected from transportation systems. (data set list)
• Two standard data modeling problems (imputation & prediction).
• A number of machine learning solutions (baseline models & newly proposed solutions).
• Well-documented Python implementation on Jupyter Notebook (mainly relied on Numpy).
• Offering ideas for spatiotemporal traffic data modeling.

Credit: Xinyu Chen would like to thank the Institute for Data Valorisation (IVADO) for providing the PhD Excellence Scholarship to support this open-source project.