ST-MTM (KDD 2025)

Hyunwoo Seo, Chiehyeon Lim

This repository provides the official implementation of ST-MTM from the paper ST-MTM: Masked Time Series Modeling with Seasonal-Trend Decomposition for Time Series Forecasting.

Requirements

• Python 3.9.0
• torch==2.0.1
• numpy==1.24.3
• pandas==1.5.3
• scikit-learn==1.2.2
• matplotlib==3.7.1
• tensorboardX==2.6.2.2

Dependencies can be installed using the following command:

pip install -r requirements.txt

Getting Started

1. Prepare Data

All benchmark datasets can be obtained from Google Drive, and arrange the folder as:

ST-MTM/
|-- datasets/
    |-- ETTh1.csv
    |-- ETTh2.csv
    |-- ETTm1.csv
    |-- ETTm2.csv
    |-- Weather.csv
    |-- Electricity.csv
    |-- Exchange.csv
    |-- national_illness.csv
    |-- solar_AL.txt
    |-- PEMS08/
        |-- PEMS08.npz

2. Experimental reproduction

• We provide the scripts for pre-training and finetuning for each dataset with the best hyper-parameters in our experiment at ./scripts/.

2-1. Pre-training

Pre-training ST-MTM for each dataset can be implemented through the provided scripts in ./scripts/pretrain/. For example, to pre-train ST-MTM for the ETTh1 dataset:

bash scripts/pretrain/ETTh1.sh

2-2. Fine-tuning

After pre-training ST-MTM for the dataset, fine-tuning ST-MTM for forecasting across various lengths can be implemented through the provided scripts in ./scripts/finetune/. For example, to fine-tune ST-MTM for the ETTh1 dataset:

bash scripts/finetune/ETTh1.sh

2-3. Pre-training and fine-tuning at once

To implement pre-training and fine-tuning sequentially, the scripts in ./scripts/. For example, to perform both steps at once for the electricity dataset:

bash scripts/run_electricity.sh

Complete Results of Multivariate Time Series Forecasting

We present the complete experimental results for multivariate time series forecasting, including performance across all prediction lengths.

Contact

If you have any questions or concerns. please contact ta57xr@unist.ac.kr or submit an issue.

Citation

If you find this repo useful in your research, please consider citing our paper as follows: