A Researcher-Friendly Framework for Time Series Analysis.

Train Any Model on Any Dataset.

---

This is also the official repository for the following paper:

• Learning Recursive Multi-Scale Representations for Irregular Multivariate Time Series Forecasting (ICLR 2026) [OpenReview] [arXiv]

  @inproceedings{li_LearningRecursiveMultiScale_2026,
      author = {Li, Boyuan and Liu, Zhen and Luo, Yicheng  and Ma, Qianli},
      booktitle = {International Conference on Learning Representations},
      title = {Learning Recursive Multi-Scale Representations for Irregular Multivariate Time Series Forecasting},
      year = {2026}
  }
  

• HyperIMTS: Hypergraph Neural Network for Irregular Multivariate Time Series Forecasting (ICML 2025) [poster] [OpenReview] [arXiv]

  @inproceedings{li_HyperIMTSHypergraphNeural_2025,
      author = {Li, Boyuan and Luo, Yicheng and Liu, Zhen and Zheng, Junhao and Lv, Jianming and Ma, Qianli},
      booktitle = {Forty-Second International Conference on Machine Learning},
      title = {HyperIMTS: Hypergraph Neural Network for Irregular Multivariate Time Series Forecasting},
      year = {2025}
  }
  

1. ✨ Hightlighted Features

• Extensibility: Adapt your model/dataset once, train almost any combination of "model" $\times$ "dataset" $\times$ "loss function".
• Compatibility: Accept models with any number/type of arguments in forward; Accept datasets with any number/type of return values in __getitem__; Accept tailored loss calculation for specific models.
• Maintainability: No need to worry about breaking the training codes of existing models/datasets/loss functions when adding new ones.
• Reproducibility: Minimal library dependencies for core components. Try the best to get rid of fancy third-party libraries (e.g., PyTorch Lightning, EasyTorch).
• Efficiency: Multi-GPU parallel training; Python built-in logger; structured experimental result saving (json)...
• Transferability: Even if you don't like our framework, you can still easily find and copy the models/datasets you want. No overwhelming encapsulation.

2. 🧭 Documentation

Checkout the new documentation website.

3. 🤖 Models

51 models, covering regular, irregular, pretrained, and traffic models, have been included in PyOmniTS, and more are coming.

Model classes can be found in models/, and their dependencies can be found in layers/

• ✅: supported
• ❌: not supported
• '-': not implemented
• MTS: regularly sampled multivariate time series
• IMTS: able to handle irregularly sampled multivariate time series

Model	Venue	Type	Forecasting	Classification	Imputation
Ada-MSHyper	NeurIPS 2024	MTS	✅	✅	✅
APN	AAAI 2026	IMTS	✅	-	-
Autoformer	NeurIPS 2021	MTS	✅	✅	✅
Scaleformer	ICLR 2023	MTS	✅	-	✅
BigST	VLDB 2024	MTS	✅	✅	✅
Crossformer	ICLR 2023	MTS	✅	✅	✅
CRU	ICML 2022	IMTS	✅	❌	✅
DLinear	AAAI 2023	MTS	✅	✅	✅
ETSformer	arXiv 2022	MTS	✅	✅	✅
FEDformer	ICML 2022	MTS	✅	✅	✅
FiLM	NeurIPS 2022	MTS	✅	✅	✅
FourierGNN	NeurIPS 2023	MTS	✅	✅	✅
FreTS	NeurIPS 2023	MTS	✅	✅	✅
GNeuralFlow	NeurIPS 2024	IMTS	✅	❌	✅
GraFITi	AAAI 2024	IMTS	✅	✅	✅
GRU-D	Scientific Reports 2018	IMTS	✅	✅	✅
HD-TTS	ICML 2024	IMTS	✅	-	✅
Hi-Patch	ICML 2025	IMTS	✅	✅	✅
higp	ICML 2024	MTS	✅	✅	✅
HyperIMTS	ICML 2025	IMTS	✅	-	✅
Informer	AAAI 2021	MTS	✅	✅	✅
iTransformer	ICLR 2024	MTS	✅	✅	✅
Koopa	NeurIPS 2023	MTS	✅	❌	✅
Latent_ODE	NeurIPS 2019	IMTS	✅	❌	✅
Leddam	ICML 2024	MTS	✅	✅	✅
LightTS	arXiv 2022	MTS	✅	✅	✅
Mamba	Language Modeling 2024	MTS	✅	✅	✅
MICN	ICLR 2023	MTS	✅	✅	✅
MOIRAI	ICML 2024	Any	✅	-	✅
mTAN	ICLR 2021	IMTS	✅	✅	✅
NeuralFlows	NeurIPS 2021	IMTS	✅	❌	✅
NHITS	AAAI 2023	MTS	✅	-	✅
Nonstationary Transformer	NeurIPS 2022	MTS	✅	✅	✅
PatchTST	ICLR 2023	MTS	✅	✅	✅
Pathformer	ICLR 2024	MTS	✅	-	✅
PrimeNet	AAAI 2023	IMTS	✅	✅	✅
Pyraformer	ICLR 2022	MTS	✅	✅	✅
Raindrop	ICLR 2022	IMTS	✅	✅	✅
Reformer	ICLR 2020	MTS	✅	✅	✅
ReIMTS	ICLR 2026	IMTS	✅	✅	-
SeFT	ICML 2020	IMTS	✅	✅	✅
SegRNN	arXiv 2023	MTS	✅	✅	✅
Temporal Fusion Transformer	arXiv 2019	MTS	✅	-	-
TiDE	TMLR 2023	MTS	✅	✅	✅
TimeCHEAT	AAAI 2025	MTS	✅	✅	✅
TimeMixer	ICLR 2024	MTS	✅	✅	✅
TimesNet	ICLR 2023	MTS	✅	✅	✅
tPatchGNN	ICML 2024	IMTS	✅	✅	✅
Transformer	NeurIPS 2017	MTS	✅	✅	✅
TSMixer	TMLR 2023	MTS	✅	✅	✅
Warpformer	KDD 2023	IMTS	✅	✅	✅

4. 💾 Datasets

Dataest classes are put in data/data_provider/datasets, and dependencies can be found in data/dependencies:

11 datasets, covering regular and irregular ones, have been included in PyOmniTS, and more are coming.

• ✅: supported
• ❌: not supported
• '-': not implemented
• MTS: regularly sampled multivariate time series
• IMTS: irregularly sampled multivariate time series

Dataset	Type	Field	Forecasting
ECL	MTS	electricity	✅
ETTh1	MTS	electricity	✅
ETTm1	MTS	electricity	✅
Human Activity	IMTS	biomechanics	✅
ILI	MTS	healthcare	✅
MIMIC III	IMTS	healthcare	✅
MIMIC IV	IMTS	healthcare	✅
PhysioNet'12	IMTS	healthcare	✅
Traffic	MTS	traffic	✅
USHCN	IMTS	weather	✅
Weather	MTS	weather	✅

Datasets for classification and imputation have not released yet.

5. 📉 Loss Functions

The following loss functions are included under loss_fns/:

Loss Function	Task	Note
CrossEntropyLoss	Classification	-
MAE	Forecasting/Imputation	-
ModelProvidedLoss	-	Some models prefer to calculate loss within forward(), such as GNeuralFlows.
MSE_Dual	Forecasting/Imputation
MSE	Forecasting/Imputation	-

6. 🚧 Roadmap

PyOmniTS is continously evolving:

• More tutorials.
• Classification support in core components.
• Imputation support in core components.
• Optional python package management via uv.

Yet Another Code Framework?

We encountered the following problems when using existing ones:

• Argument & return value chaos for models' forward():

  Different models usually take varying number and shape of arguments, especially ones from different domains. Changes to training logic are needed to support these differences.

• Return value chaos for datasets' __getitem__():

  datasets can return a number of tensors in different shapes, which have to be aligned with arguments of models' forward() one by one. Changes to training logic are also needed to support these differences.

• Argument & return value chaos for loss functions' forward():

  loss functions take different types of tensors as input, require aligning with return values from models' forward().

• Overwhelming dependencies:

  some existing pipelines use fancy high-level packages in building the pipeline, which can lower the flexibility of code modification.

Contributors

Ladbaby 💻 🐛

Acknowledgement

• Time Series Library: Models and datasets for regularly sampled time series are mostly adapted from it.
• BasicTS: Documentation design reference.
• Google Gemini: Icon creation.