Predictability-Aware Compression and Decompression Framework (PCDF)

📢 News:
This work has been accepted for publication at The Web Conference 2026 (WWW 2026). 📄 arXiv Paper: https://arxiv.org/abs/2506.00614

This repository contains the official implementation of our submission:

Predictability-Aware Compression and Decompression Framework for Multichannel Time Series Data with Latent Seasonality

📦 Requirements

The code is written in Python and requires the following packages:

• torch~=2.6.0+cu126
• numpy~=2.1.2
• pandas~=2.2.3
• tqdm~=4.67.1
• meteostat~=1.6.8
• matplotlib~=3.10.0
• scipy~=1.15.1
• scikit-learn~=1.6.1
• psutil~=5.9.0

You can install all dependencies via:

pip install -r requirements.txt

📁 Project Structure:

PCDF/
├── Designed methods.py # Our proposed the compression-decompression framework
├── Comparative methods.py # Baseline comparison methods
├── Data/ # Input datasets
├── Main.py # Entry point for running experiments
└── README.md # Project documentation

🚀 How to Run

To reproduce the experiments:

1. Select and prepare your dataset (place it under the data/ directory).
2. Modify relevant parameters in main.py according to the paper.
3. Run the experiment with: python main.py
4. The output will include the MSE and runtime of both the proposed and comparative methods.

📊 Datasets

We evaluate our method on the following publicly available multivariate time series datasets:

• NYC Taxi
• DC Bike
• Electricity Load Diagrams
• Solar Energy
• Gas Sensor Array Drift
• Weather (Meteostat)

⚠️ Note: Some of the datasets exceed GitHub's file size limit and are not included in this repository.
Please download them from their official websites using the links provided in the paper or in the dataset section below.