The provided function gaussian_kde_gpu() is a simplified version of Scipy's gaussian_kde. It does not support weights and only uses the default Scott's Rule for bandwidth estimation. However, it leverages the computative power of a CUDA supported GPU via Numba. Given the random variable p it will estimate the probability density function (PDF) at query point q in the following way:
from gaussian_kde_gpu import gaussian_kde_gpu
density = gaussian_kde_gpu(p, q)
For more details see tutorial.
- Numpy
- Numba