This code is the official PyTorch implementation of our NIPS'25 paper: Enhancing Time Series Forecasting through Selective Representation Spaces: A Patch Perspective.
If you find this project helpful, please don't forget to give it a ⭐ Star to show your support. Thank you!
🚩 News (2025.9) Our paper has been accepted as a Spotlight poster in NeurIPS 2025.
In this paper, we pioneer the exploration of constructing a selective representation space to flexibly include the information beneficial for forecasting. Specifically, we propose the Selective Representation Space (SRS) module, which utilizes the learnable Selective Patching and Dynamic Reassembly techniques to adaptively select and shuffle the patches from the contextual time series, aiming at fully exploiting the information of contextual time series to enhance the forecasting performance of patch-based models. To demonstrate the effectiveness of SRS module, we propose a simple yet effective SRSNet consisting of SRS and an MLP head, which achieves state-of-the-art performance on real-world datasets from multiple domains.
The important components of the SRS Module: (1) Selective Patching; (2) Dynamic Reassembly ; (3) Adaptive Fusion
Important
this project is fully tested under python 3.8, it is recommended that you set the Python version to 3.8.
- Requirements
Given a python environment (note: this project is fully tested under python 3.8), install the dependencies with the following command:
pip install -r requirements.txt- Data preparation
You can obtained the well pre-processed datasets from Google Drive. Then place the downloaded data under the folder ./dataset.
- Train and evaluate model
- To see the model structure of SRSNet, click here.
- We provide all the experiment scripts for SRSNet and other baselines under the folder
./scripts/multivariate_forecast. For example you can reproduce all the experiment results as the following script:
sh ./scripts/multivariate_forecast/ETTh1_script/SRSNet.shExtensive experiments on 8 real-world datasets demonstrate that SRSNet achieves state-of-the-art~(SOTA) performance.
If you find this repo useful, please cite our paper.
@inproceedings{wu2025srsnet,
title = {Enhancing Time Series Forecasting through Selective Representation Spaces: A Patch Perspective},
author = {Wu, Xingjian and Qiu, Xiangfei and Cheng, Hanyin and Li, Zhengyu and Hu, Jilin and Guo, Chenjuan and Yang, Bin},
booktitle = {NeurIPS},
year = {2025}
}
If you have any questions or suggestions, feel free to contact:
Or describe it in Issues.
