aptx1231 / start

[ICDE2023] A PyTorch implementation of Self-supervised Trajectory Representation Learning with Temporal Regularities and Travel Semantics Framework (START).

Hi, welcome to my profile 👋

• 🤔 I am Jiawei Jiang, a graduate student at the School of Computer Science and Engineering, Beihang University, China (UTC+8).
• 💡 I am also a member of Baidu PaddlePaddle Developers Experts (PPDE), a reviewer of IEEE Transactions on Neural Networks and Learning Systems (IEEE TNNLS) and IEEE Transactions on Knowledge and Data Engineering (IEEE TKDE).
• 🌱 My research interests include Spatial-Temporal Data Mining, Graph Neural Networks, and Representation Learning.
• 💪 I am the core developer of LibCity, an open-source library for Urban Spatial-temporal Data Mining. The paper LibCity: An Open Library for Traffic Prediction has been accepted for the 29th [ACM SIGSPATIAL 2021] conference. Welcome to our [Code], [Paper], and [Website] for more details. We also publish a full paper titled LibCity: A Unified Library Towards Efficient and Comprehensive Urban Spatial-Temporal Prediction, which provides more details. Statistics: .
• 💪 We publish a paper titled Unified Data Management and Comprehensive Performance Evaluation for Urban Spatial-Temporal Prediction [Experiment, Analysis & Benchmark], including (1) a Unified Storage Format for urban spatial-temporal data, (2) a Technical Development Roadmap for urban spatial-temporal prediction models, (3) Extensive Experiments and Performance Evaluation using 18 models and 20 datasets. [Paper].
• 🚀 Our paper titled Self-supervised Trajectory Representation Learning with Temporal Regularities and Travel Semantics has been accepted for the 39th IEEE International Conference on Data Engineering [ICDE 2023] conference (CCF A). [Paper], [Code]. Rank: 1/ 6.
• 🚀 Our paper titled PDFormer: Propagation Delay-aware Dynamic Long-range Transformer for Traffic Flow Prediction has been accepted for the 37th AAAI Conference on Artificial Intelligence [AAAI 2023] conference (CCF A). [Paper], [Code]. Rank: 1/ 4.
• 🎉 We rank 11th in the Regular track of the Baidu KDD Cup 2022 for Spatial Dynamic Wind Power Forecasting Challenge. The paper has been accepted for [ACM SIGKDD Workshop 2022] conference. [Paper] [Code].
• 🎉 We rank 6th in the ACM RecSys Challenge 2023 for Advertising Installation Prediction Challenge. [Paper], [Code].
• 🎉 We rank 3th in the iFLYTEK A.I. Developer Competition 2022 for Urban Road Traffic Flow Forecasting Challenge. [Code].
• ⚡ Languages and environments: Python, Pytorch.
• 😇 Pronouns: He/Him/His.
• 🎓 Link to my Google Scholar and Blog.
• 📫 Reach me by email: [email protected].

start's Issues

About travel time estimation

About Temporal Shifting Strategy Implementation

I have been studying your work and I am particularly interested in the code implementation of the augmentation strategies mentioned in the paper.

While exploring the publicly available code, I noticed that the specific code for the mentioned augmentation strategies is not publicly accessible. However, through reverse engineering on the publicly available data, I discovered that the paper states the selection of subsequences for the shift operation with a probability of 0.15. During the reverse engineering process, we found that there are multiple subsequences for the shift operation on a single road (inferred from the tau g of the shift). We are interested in obtaining a more detailed understanding of this aspect of the implementation.

If possible, could you kindly provide us with the detailed settings and, if available, the code related to these augmentation strategies? We would greatly appreciate any assistance you can provide in this regard.

[Request] Preprocess code of Geolife or raw_data of Geolife used in the "Transfer Model Across Datasets" Experiment

Hello aptx1231, this is excellent work, and thank you very much for sharing the data and code!

In addition, I would like to inquire whether it is possible to disclose the preprocessing procedures for the Geolife dataset or the raw_data of Geolife used in your experiments in the "Transfer Model Across Datasets" section (as shown below).

Thanks a lot!

Creating my own dataset

Hi @aptx1231 ,
Thanks for this famous research. I am currently researching on the same field only. I have few questions regarding your research.

1. I want to understand that how you had created the dataset ? Specially the path column in the porto's raw data. H

2. How you created the porto_eval_enhancedbytrim.csv and porto_eval_enhancedbyshift.csv ? How is it important for the code ?

3. How did you exported porto_roadmap_edge.geo and porto_roadmap_edge.rel ? Can you please elaborate on it ?

[Request] Preprocessed data of Geolife for trajectory classification experiment

Hi, aptx1231!

I have been runing the map matching task on raw Geolife data to generate the required format for START input using HMMM. However, the progress has been slower than anticipated, nearly 20 hours passed and completion remains elusive.

In light of this, I was wondering if you could kindly share the preprocessed Geolife data that could be directly utilized as input for your trajectory classifying experiment. Your assistance would be greatly appreciated!