Coreset-KR

This repository contains the implementation of "A Topology-Preserving Coreset for Kernel Regression in Scientific Visualization", accepted to PacificVis 2026 TVCG Journal track.

coreset_kr.py implements coreset for kernel regression based on gradient descent optimization for 2D scalar field data. The script constructs and optimizes coresets (compressed representations of large datasets) that preserve the kernel regression behavior of the original data.

USAGE

The dataset GTOPO-ME is used here as an example. You need to replace it with your dataset.

Create data path

Run the following commands. Need to run this only once.

cd data/GTOPO-ME/
mkdir VTKfiles

Python Environment

Create a virtual environment and install required dependencies:

# Install required packages
pip install numpy torch vtk pyevtk scikit-learn pandas

Required packages:

• numpy: Numerical computations
• torch: PyTorch for GPU acceleration and optimization
• vtk: VTK for reading/writing visualization data
• pyevtk: VTK file writing utilities
• scikit-learn: KDTree for neighbor search
• pandas: CSV file reading

Note: For GPU support, ensure you have CUDA installed and install the appropriate PyTorch version with CUDA support from pytorch.org.

Optimization

To optimize coreset for kernel regression using gradient descent, use command

python src/coreset_kr.py <dataName> <sigma> <learning_rate> <run_KR_S> <run_random> <randomSample> <scalarfield>

Parameters:

• dataName: Name of the dataset (e.g., 'GTOPO-ME')
• sigma: Kernel bandwidth parameter
• learning_rate: Learning rate for optimization
• run_opt: Boolean flag to run coreset optimization
• run_random: Boolean flag to run GR baseline
• run_randomSample: Boolean flag to run RS baseline
• scalarfield: Name of the scalar field in VTK files (e.g., 'scalar')

Example

python src/coreset_kr.py GTOPO-ME 5 1 True True True scalar

CITATION

If you found this work useful, please consider citing it as

@article{lyu2026coreset,
  title={A Topology-Preserving Coreset for Kernel Regression in Scientific Visualization},
  author={Lyu, Weiran and Gorski, Nathaniel and Phillips, Jeff M. and Wang, Bei},
  journal={IEEE Transactions on Visualization and Computer Graphics},
  year={2026}
}

Notes

• The script supports both CPU and GPU execution
• Code comments are still in progress