Skip to main content

We implemented nadaraya waston kernel density and kernel conditional probability estimator using cuda through cupy. It is much faster than cpu version but it requires GPU with high memory.

Project description


Kernel density estimator using Nadaraya-Watson with GPU (CUDA)


Chen Chen

Running Environment Setup

You have to have cupy installed to be able to use GPU!! See

Similar to scipy.kde_gaussian and statsmodels.nonparametric.kernel_density.KDEMultivariateConditional, we implemented nadaraya waston kernel density and kernel conditional probability estimator using cuda through cupy. However, it is much faster than cpu version and it maximise the use of GPU memory.

  1. Make a folder with name "data" in current directory. Then copy ORL and CroppedYaleB dataset inside. Please make sure you have the following file tree structure:
    |--- kde_gpu\
    ***|--- \
    ***|--- \
    ***|--- \
    *|--- \
    *|--- \
    *|--- \

  2. Install kde_gpu with following command: (Please use pip3 if the default python in your computer is python2)

    $ pip install -e .

This command will run where we specify the dependencies required to run nmf. The dependencies we require are:


Please note that if the version number of installed package in your machine is lower than the stated version number, pip will uninstall your out-of-date package and install the one with version number greater than or equal to the stated one in



#import kernel_smoothing
from scipy import stats
import pandas as pd
import cupy as cp
import numpy as np
import time

rv = stats.expon(0,1)

x = rv.rvs(size=10000)

density_real = rv.pdf(x)



df = pd.DataFrame({'x1':x,'kde_scipy':kde_scipy,
                   'kde_cupy':cp.asnumpy(kde_cupy).squeeze(),'real density':density_real})

df['scipy_mean_absolute_error']=np.abs(df['kde_scipy']-df['real density'])
df['cupy_mean_absolute_error']=np.abs(df['kde_cupy']-df['real density'])

rv = stats.truncnorm(-3,2,30,10)
x = cp.asarray(rv.rvs(nsample))
ycondx = cp.asarray(cp.random.rand(nsample))
y = 10*(ycondx-0.5)+x

cdf_conditional_real = ycondx
df = pd.DataFrame({'y':cp.asnumpy(y),'x':cp.asnumpy(x),'real density':cp.asnumpy(cdf_conditional_real)})

df['nadaraya watson']= kernel_smoothing_ecdf(y,x)
df['nw_error']=np.abs(df['nadaraya watson']-df['real density'])

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Files for kde-gpu, version 0.1.0
Filename, size File type Python version Upload date Hashes
Filename, size kde_gpu-0.1.0-py3-none-any.whl (9.0 kB) File type Wheel Python version py3 Upload date Hashes View
Filename, size kde_gpu-0.1.0.tar.gz (4.4 kB) File type Source Python version None Upload date Hashes View

Supported by

Pingdom Pingdom Monitoring Google Google Object Storage and Download Analytics Sentry Sentry Error logging AWS AWS Cloud computing DataDog DataDog Monitoring Fastly Fastly CDN DigiCert DigiCert EV certificate StatusPage StatusPage Status page