Figure 1. Overall approach of DeepTime.
Official PyTorch code repository for the DeepTime paper. Check out our blog post!
- DeepTime is a deep time-index based model trained via a meta-optimization formulation, yielding a strong method for time-series forecasting.
- Experiments on real world datases in the long sequence time-series forecasting setting demonstrates that DeepTime achieves competitive results with state-of-the-art methods and is highly efficient.
Dependencies for this project can be installed by:
pip install -r requirements.txt
To get started, you will need to download the datasets as described in our paper:
- Pre-processed datasets can be downloaded from the following links, Tsinghua Cloud or Google Drive, as obtained from Autoformer's GitHub repository.
- Place the downloaded datasets into the
storage/datasets/
folder, e.g.storage/datasets/ETT-small/ETTm2.csv
.
We provide some scripts to quickly reproduce the results reported in our paper. There are two options, to run the full hyperparameter search, or to directly run the experiments with hyperparameters provided in the configuration files.
Option A: Run the full hyperparameter search.
- Run the following command to generate the experiments:
make build-all path=experiments/configs/hp_search
. - Run the following script to perform training and evaluation:
./run_hp_search.sh
(you may need to runchmod u+x run_hp_search.sh
first).
Option B: Directly run the experiments with hyperparameters provided in the configuration files.
- Run the following command to generate the experiments:
make build-all path=experiments/configs/ETTm2
. - Run the following script to perform training and evaluation:
./run.sh
(you may need to runchmod u+x run.sh
first).
Finally, results can be viewed on tensorboard by running tensorboard --logdir storage/experiments/
, or in
the storage/experiments/experiment_name/metrics.npy
file.
We conduct extensive experiments on both synthetic and real world datasets, showing that DeepTime has extremely competitive performance, achieving state-of-the-art results on 20 out of 24 settings for the multivariate forecasting benchmark based on MSE.
Further details of the code repository can be found here. The codebase is structured to generate experiments from
a .gin
configuration file based on the build.variables_dict
argument.
- First, build the experiment from a config file. We provide 2 ways to build an experiment.
- Build a single config file:
make build config=experiments/configs/folder_name/file_name.gin
- Build a group of config files:
make build-all path=experiments/configs/folder_name
- Build a single config file:
- Next, run the experiment using the following command
Alternatively, the first step generates a command file found in
python -m experiments.forecast --config_path=storage/experiments/experiment_name/config.gin run
storage/experiments/experiment_name/command
, which you can use by the following command,make run command=storage/experiments/experiment_name/command
- Finally, you can observe the results on tensorboard
or view the
tensorboard --logdir storage/experiments/
storage/experiments/deeptime/experiment_name/metrics.npy
file.
The implementation of DeepTime relies on resources from the following codebases and repositories, we thank the original authors for open-sourcing their work.
- https://github.com/ElementAI/N-BEATS
- https://github.com/zhouhaoyi/Informer2020
- https://github.com/thuml/Autoformer
Please consider citing if you find this code useful to your research.
@InProceedings{pmlr-v202-woo23b, title = {Learning Deep Time-index Models for Time Series Forecasting}, author = {Woo, Gerald and Liu, Chenghao and Sahoo, Doyen and Kumar, Akshat and Hoi, Steven}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {37217--37237}, year = {2023}, editor = {Krause, Andreas and Brunskill, Emma and Cho, Kyunghyun and Engelhardt, Barbara and Sabato, Sivan and Scarlett, Jonathan}, volume = {202}, series = {Proceedings of Machine Learning Research}, month = {23--29 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v202/woo23b/woo23b.pdf}, url = {https://proceedings.mlr.press/v202/woo23b.html} }