This repo is the official implementation for Scale-Aware Neural Architecture Search for Multivariate Time Series Forecasting.

1.1 The framework of SNAS4MTF

2 Prerequisites

• Python 3.6.12
• PyTorch 1.2.0
• math, sklearn, numpy

3 Datasets

3.1 METR-LA

This dataset is collected by the Los Angeles Metropolitan Transportation Authority and contains the average traffic speed measured by 207 loop detectors on the highways of Los Angeles County between March 2012 and June 2012.

3.2 PEMS-BAY

The raw data is in http://pems.dot.ca.gov. This dataset is collected by California Transportation Agencies and contains the average traffic speed measured by 325 sensors in the Bay Area between January 2017 and May 2017.

4 Running

4.1 Install all dependencies listed in prerequisites

4.2 Download the dataset

4.3 Neural Architecture Search

# Neural Architecture Search on PEMS_BAY
 python search.py --config config/PEMS_BAY_para.yaml |& tee logs/search_PEMS_BAY.log
 # Neural Architecture Search on METR_LA
 python search.py --config config/METR_LA_para.yaml |& tee logs/search_METR_LA.log

4.4 Training

# Train on PEMS_BAY
python train.py --config config/PEMS_BAY_para.yaml  |& tee logs/train_PEMS_BAY.log
# Train on METR-LA
python train.py --config config/METR_LA_para.yaml |& tee logs/train_METR_LA.log

4.5 Evaluating

# Evaluate on PEMS_BAY
python test.py --config config/PEMS_BAY_para.yaml |& tee logs/test_PEMS_BAY.log
# Evaluate on METR-LA
python test.py --config config/METR_LA_para.yaml |& tee logs/test_METR_LA.log

5 Concat

If you have any questions, please feel free to contact zongjiangshang@cs.zju.edu.cn

6 Other works

📝 Our other works are shown as follows：

Single-step forecasting: Chen L, Chen D, Shang Z, et al. Multi-scale adaptive graph neural network for multivariate time series forecasting. TKDE, 2023, 35(10): 10748-10761. Code Link

Long-range & short-range & ultra-long-range time series forecasting: Shang Z, Chen L, Wu B, et al. Ada-MSHyper: Adaptive multi-scale hypergraph transformer for time series forecasting. NeurIPS, 2024. Code Link

Long-range time series forecasting: Shang Z, Chen L, Wu B, et al. MSHyper: Multi-scale hypergraph transformer for long-range time series forecasting. arXiv, 2024: arXiv: 2401.09261. Code Link

7 Citation

Please cite the following paper if you use the code in your work:

@article{chen2021scale,
  title={Scale-aware neural architecture search for multivariate time series forecasting},
  author={Chen, Donghui and Chen, Ling and Shang, Zongjiang and Zhang, Youdong and Wen, Bo and Yang, Chenghu},
  journal={ACM Transactions on Knowledge Discovery from Data},
  year={2024},
  publisher={ACM New York, NY}
}