parallelkdepy 1.0.0

Python wrapper of ParallelKDE.jl

ParallelKDEpy: Python wrapper for ParallelKDE.jl

High performance implementation of a parallel kernel density estimation algorithm described in [Sustay Martinez et al. (2025)]. The algorithm is specially suited for high-dimensional data, with CPU/CUDA support.

Quick links:

• 📑 Docs: stable | dev
• 📇 Citing
• Julia package: ParallelKDE.jl

Installation

pip install parallelkdypy

The wrapper will handle installing Julia and ParallelKDE.jl the first time you import it. No additional setup is required.

Quick Start

import numpy as np
import parallelkdepy as pkde

# Assume 'data' is a 2D array of points for which we want to estimate the density
data = np.random.randn(10000, 1)

density_estimation = pkde.DensityEstimation(data, grid=True, device="cpu")
density_estimation.estimate_density("gradepro")

density estimated = density_estimation.get_density()

See the documentation for more details on how to use the package.

Features

Currently, there are two estimators available:

• "gradepro": As described in [Sustay Martinez et al. (2025)], this estimator is designed for high-dimensional data and can be run on both CPU and GPU.
• "rot": Implements the rules of thumb (Silverman and Scott) for bandwidth selection. It makes use of some of the routines from :gradepro to evaluate the density on a grid.

Julia Package

Please keep in mind that this package is a wrapper around the Julia package ParallelKDE.jl. For more information on the Julia package, please refer to its documentation.

Citing

Please cite the following papers when using ParallelKDEpy in your work:

Known Issues

List of main known issues:

• Issue #5: Segmentation faults for CUDA implementations