OCE-TS

Beyond MSE: Ordinal Cross-Entropy for Probabilistic Time Series Forecasting

OCE-TS is a novel probabilistic time series forecasting framework that replaces traditional Mean Squared Error (MSE) regression with an Ordinal Cross-Entropy (OCE) classification paradigm. The method provides more stable optimization, improved robustness, and inherent uncertainty estimation across multiple benchmark datasets.

---

🚀 Features

• Probabilistic forecasting with uncertainty estimation
• Ordinal Cross-Entropy (OCE) loss to preserve temporal ordering
• Robust to noise and outliers
• Supports multiple datasets (ETTh, ETTm, Exchange, Weather, ILI, etc.)
• Lightweight backbone based on DLinear

---

📦 Environment & Requirements

This project was implemented in the following environment:

• Python 3.8.20
• PyTorch 2.0.0 (CUDA 11.8)
• GPU: NVIDIA RTX 4060
• CPU: Intel i7-14700F
• RAM: 16 GB

Install dependencies:

pip install -r requirements.txt

---

🏃 Usage

Run (Train + Test)

Running the script below will automatically perform both training and testing:

python run_longExp.py

MyLinear/
│── Da/ # Data augmentation or auxiliary data
│── data_provider/ # Data loading and preprocessing
│── dataset/ # Dataset files and utilities
│── exp/ # Experiment manager (training/testing)
│── layers/ # Model layers
│── logs/ # Training logs
│── loss/ # Loss functions (OCE, CE, Gaussian, etc.)
│── models/ # Forecasting models (DLinear, OCE-TS modules)
│── results/ # Result figures, MAE/MSE plots, tables
│── scripts/ # Helper scripts / run scripts
│── test_results/ # Saved test outputs
│── utils/ # Utility functions (Gaussian, noise, lookback, etc.)
│
│── distribution.py
│── Gaussian.py
│── lookback.py
│── loss_compare.py
│── ModelComparisonPlot.py
│── noise.py
│── OCE vs CE.py
│── Parameter.py
│── Probability Distribution Comparison.py
│── run_longExp.py # Main entry script (train+test)
│── sh.py
│── result.txt
│── *.png # Visualization images


---

## 📊 Results

Our method (OCE-TS) consistently outperforms strong baselines such as  
Autoformer, DLinear, iTransformer, TimeXer, and TimeBridge across multiple datasets.

For detailed experimental results, tables, and comparisons,  
**please refer to the paper**.


## 📚 Citation

If you use this repository, please cite:

```bibtex
@misc{wang2025mseordinalcrossentropyprobabilistic,
	title={Beyond MSE: Ordinal Cross-Entropy for Probabilistic Time Series Forecasting}, 
	author={Jieting Wang and Huimei Shi and Feijiang Li and Xiaolei Shang},
	year={2025},
	eprint={2511.10200},
	archivePrefix={arXiv},
	primaryClass={cs.LG},
	url={https://arxiv.org/abs/2511.10200}, 
}

---

📄 License

MIT License

---