The repo is the official implementation for the paper: "FilterNet: Harnessing Frequency Filters for Time Series Forecasting".
To get started, ensure you have Conda installed on your system and follow these steps to set up the environment:
conda create -n FilterNet python=3.8
conda activate FilterNet
pip install -r requirements.txt
All the datasets needed for FilterNet can be obtained from the Google Drive provided in Autoformer.
For datasets with a small number of variables, such as ETTh, ETTm, and Exchange, we recommend using PaiFilter as follows:
bash ./scripts/PaiFilter/ETTm1.sh
bash ./scripts/PaiFilter/ETTm2.sh
bash ./scripts/PaiFilter/ETTh2.sh
For datasets with a large number of variables such as ECL, Traffic, and weather, it is recommended to use TexFilter as follows:
bash ./scripts/PaiFilter/ECL.sh
bash ./scripts/PaiFilter/Traffic.sh
bash ./scripts/PaiFilter/Weather.sh
👉 News (2024.12): Another one of our recent works, Amplifier: Bringing Attention to Neglected Low-Energy Components in Time Series Forecasting, has been accepted by AAAI 2025.
🚩 [IJCAI 2024]: Deep Frequency Derivative Learning for Non-stationary Time Series Forecasting
🚩 [NeurIPS 2023]: Frequency-domain MLPs are more effective learners in time series forecasting
🚩 [NeurIPS 2023]: FourierGNN: Rethinking Multivariate Time Series Forecasting from a Pure Graph Perspective
🚩 [arXiv]: A Survey on Deep Learning based Time Series Analysis with Frequency Transformation
We appreciate the following GitHub repositories for providing valuable code bases and datasets:
https://github.com/wanghq21/MICN
https://github.com/thuml/TimesNet
https://github.com/aikunyi/FreTS
https://github.com/VEWOXIC/FITS
https://github.com/plumprc/RTSF
https://github.com/cure-lab/LTSF-Linear
https://github.com/zhouhaoyi/Informer2020
https://github.com/thuml/Autoformer
https://github.com/ant-research/Pyraformer
https://github.com/MAZiqing/FEDformer
https://github.com/yuqinie98/PatchTST