Skip to content

Latest commit

 

History

History
53 lines (32 loc) · 2.05 KB

README.md

File metadata and controls

53 lines (32 loc) · 2.05 KB

Preformer

This repository contains the pytorch code for the 2023 ICASSP paper "Preformer: Predictive Transformer with Multi-Scale Segment-wise Correlations for Long-Term Time Series Forecasting”.

Model

The core MSSC module:

Main Results

Acknowledgment

This repository uses some code from Autoformer and Informer. Thanks to the authors for their work!

Get Started

Environment

Install Python 3.6, PyTorch 1.9.0.

Data

We use all the datasets provided by Autoformer directly, so you can download them from Google drive provided in Autoformer. All the datasets are well pre-processed and can be used easily. After downloading, put these dataset files (such as ETTh1.csv) in the ./data/ folder.

Train

python run.py --is_training 1 --root_path ./data/ --data_path ETTh1.csv --model_id ETTh1_96_48 --model Preformer --data ETTh1 --features M --seq_len 96 --label_len 48 --pred_len 48 --e_layers 2 --d_layers 1 --factor 4 --enc_in 7 --dec_in 7 --c_out 7 --des 'Exp' --itr 1 --n_heads 4 --d_model 32 --d_ff 128

Test

python run.py --is_training 0 --root_path ./data/ --data_path ETTh1.csv --model_id ETTh1_96_48 --model Preformer --data ETTh1 --features M --seq_len 96 --label_len 48 --pred_len 48 --e_layers 2 --d_layers 1 --factor 4 --enc_in 7 --dec_in 7 --c_out 7 --des 'Exp' --itr 1 --n_heads 4 --d_model 32 --d_ff 128

Main Results

Main Results

Citation

@inproceedings{du2022preformer,
  title={Preformer: Predictive Transformer with Multi-Scale Segment-wise Correlations for Long-Term Time Series Forecasting},
  author={Du, Dazhao and Su, Bing and Wei, Zhewei},
  booktitle={ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
  year={2023},
  organization={IEEE}
}