Take a sneak peek at the new NIST.gov and let us know what you think!
(Please note: some content may not be complete on the beta site.).

View the beta site
NIST logo
Bookmark and Share



 gppois is a Bayesian tool for learning about a smooth function based on noisy measurements of its values. Users train a model of the true function on the noisy datapoints, then use that model to make predictions at whatever points they desire to know.

UPDATE (2012-09-10): gppois is now on CRAN, and even easier to install!

Important: I have moved development to github!  The latest version of the software can always be found there.  It also has an "Issues" page, where you can easily submit any problems you may have and track their status.  Please visit the github page for gppois to keep track of all the latest updates.

NOTE: this software is associated with a paper of ours, "A Bayesian approach for denoising one-dimensional data," published in the Journal of Applied Crystallography; we provide instructions for reproducing the results of that paper.


gppois quantifies uncertainty in continuous functions, given a collection of (possibly noisy) measurements of function values.

This uncertainty usually comes from two sources:

  1. Noise: the measured values differ randomly from the true values
    • Example: X-ray powder diffraction datapoints exhibit Poisson noise, or "counting statistics"
  2. Interpolation: we know the function in one region; what does that tell us about another?


Version: 0.2-1

Last Updated: 2012-07-17

Type of software: R package


Charles R. Hogg III


Requires the R software environment (available for Linux, Mac, Unix, Windows).

The following R libraries are also required or recommended; see the installation instructions.

Required libraries:

  • R.oo
Recommended libraries:
  • rgl (for visualizing 2D datasets)
  • geometry (for visualizing 2D datasets)
  • Cairo (for plotting 2D covariance functions)
  • debug (for Cholesky decomposition)