⏳✨ Conditional Diffusion Model with Nonlinear Data Transformation for Time Series Forecasting 📈🌊

August 14, 2025 · View on GitHub

ICML 2025

Welcome to the official implementation of our ICML 2025 paper:

Conditional Diffusion Model with Nonlinear Data Transformation for Time Series Forecasting

🎯 Our method blends Generative Model framework 🌀 with non-linear data transformations 🔄 to unlock state-of-the-art forecasting performance across diverse time series datasets.
Whether it’s climate 🌦, finance 💹, or energy ⚡ — this repo has you covered.

📜 Table of Contents

📚 Paper

📄 ICML 2025 — Conditional Diffusion Model with Nonlinear Data Transformation for Time Series Forecasting
📥 Read the Paper (openreview version)

📂 Datasets

Download the datasets from Google Drive link and keep them in data folder

Note : datasets link is from Autoformer paper

🛠 Installation

Clone the repo and install dependencies 🐍:

git clone this repo
cd cndiff
pip3 install -r requirements.txt

💻 Usage

To run for all the datasets

chmod +x ./scripts/run_all.sh
./scripts/run_all.sh

To run for each dataset

python3 -m scripts.run_cndiff --cfg ./< yaml file >
eg: python3 -m scripts.run_cndiff --cfg ./exchange.yaml

🤝 Citation

If you find this work useful, please cite our paper.

@inproceedings{rishiconditional,
  title={Conditional Diffusion Model with Nonlinear Data Transformation for Time Series Forecasting},
  author={Rishi, J and Mothish, GVS and Subramani, Deepak},
  booktitle={Forty-second International Conference on Machine Learning}
}

📬 Contact