How Much Can Time-related Features Enhance Time Series Forecasting?

In this repository, we present the code of "How Much Can Time-related Features Enhance Time Series Forecasting?".

Data

All the datasets are available at Autoformer: Google Drive. You only need to download electricity, ETT-small, traffic, and weather.

Environment

We implement our code in Python 3.9 and CUDA 11.7. See requirments.txt for other packages. For convenience, you can install using the following commands:

conda create -n timelinear python=3.9

pip install https://download.pytorch.org/whl/cu117_pypi_cudnn/torch-2.0.0%2Bcu117.with.pypi.cudnn-cp39-cp39-linux_x86_64.whl

pip install -r requirements.txt

Reproducibility

All the training scripts are provided in scripts/long_term_forecast. For instance, if you want to get the results for the weather dataset, you just need to run:

bash ./scripts/long_term_forecast/Weather_script/TimeLinear.sh

The default seq_len in this repository is 96. For other experimental settings, the hyperparameters that you can tune are:

--seq_len
--pred_len
--batch_size
--learning_rate
--time_feature_types    # add your time-related features, e.g., HourOfDay DayOfWeek
--rda                   # reduction rate for the first hidden layer of TimeSter
--rdb                   # reduction rate for the second hidden layer of TimeSter
--ksize                 # kernel size for the Cov1d in TimeSter
--beta                  # trade-off coefficient for the output of TimeSter and BonSter (the backbone results)

It is recommended to tune rda in {8, 4, 2, 1}, rdb in {1, 2}, ksize in {3, 5, 7}, and beta in {0.1, ..., 0.9}.

We also provide the experimental scripts for Table 3 and 4, where we combine TimeSter with state-of-the-art models. You can run the following command to reproduce the results:

bash ./scripts/long_term_forecast/ECL_script/TimePatchTST.sh

All models named Timexx (except for TimesNet) indicate we combine TimeSter with the backbone, e.g., TimePatchTST.

For convenience, you can also run the following command to reproduce all the results in Table 2:

bash ./all.sh

Results

• Checkpoints for each model will be saved in checkpoints/;
• Training log will be saved in log/;
• Prediction for the testing set will be saved in results/ (if needed) and result_long_term_forecast;
• Visualization for the results of testing set will be saved in test_results/ (if with_curve is enabled).

Our results have been stored in result_long_term_forecast and results/TimeLinear.

Acknowledgement

We are grateful for the following github repositories that provide valuable datasets and code base:

https://github.com/thuml/Autoformer

https://github.com/thuml/Time-Series-Library

https://github.com/yuqinie98/PatchTST

https://github.com/VEWOXIC/FITS

https://github.com/kwuking/TimeMixer

https://github.com/luodhhh/ModernTCN

https://github.com/ForestsKing/GLAFF

https://github.com/ACAT-SCUT/CycleNet

https://github.com/ts-kim/RevIN