Project details for pyGPs

Screenshot pyGPs 1.0

by mn - October 1, 2013, 14:12:14 CET [ Project Homepage BibTeX Download ]

view (11 today), download ( 4 today ), 3 subscriptions

Description:

pyGPs is a Python project for Gaussian process (GP) regression and classification for machine learning.

We support two libraries: pyGP_PR and pyGP_OO. pyGP_PR is currently the default download, for pyGP_OO follow this link: https://github.com/marionmari/pyGP_OO.

pyGP_PR is a procedural implementation of GPs and follows structure and functionality of the gpml matlab implementaion by Carl Edward Rasmussen and Hannes Nickisch (Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2013-01-21).

pyGP_OO is an object-oriented implemetation of GP regression and classificaion additionally supporting useful routines for the practical use of GPs, such as cross validation functionalities for evaluation as well as basic routines for iterative restarts for the GP hyperparameter optimization.

Changes to previous version:

Initial Announcement on mloss.org.

BibTeX Entry: Download
URL: Project Homepage
Supported Operating Systems: Platform Independent
Data Formats: Numpy
Tags: Classification, Regression, Gaussian Processes
Archive: download here

Other available revisons

Version Changelog Date
1.3

Changelog pyGPs v1.3

October 19th 2014

documentation updates:

  • DOC: model.fit() is now named model.getPosterior
  • DOC: typo fixed: cov.LIN changed to cov.Linear
  • DOC: removed cov.Periodic() in demos because its limited in 1-d data.
  • API file updated accordingly

structural updates:

  • removed unused ScalePrior attribute in most inference method
  • added function jitchol, which added a small jitter(instead of doing Cholesky decomposition directly) to the diagonal when the kernel matrix is ill conditioned.
  • thrown error when using periodic covariance functions for non-1d data. We also removed cov.Periodic() in demos because its limited usage.
  • combined equally spaced positions of inputs as test positions as well in plot methods to get a more accurate plotting.
  • rename model.fit() to model.getPosterior(), while model.optimize() stays the same. (since it is confusing for some users that the name fit() is not doing optimizing.)
October 20, 2014, 16:03:28
1.2

Changelog pyGPs v1.2

June 30th 2014

structural updates:

  • input target now can either be in 2-d array with size (n,1) or in 1-d array with size (n,)
  • setup.py updated
  • "import pyGPs" instead of "from pyGPs.Core import gp"
  • rename ".train()" to ".optimize()"
  • rename "Graph-stuff" to "graphExtension"
  • rename kernelOnGraph to "nodeKernels" and graphKernel to "graphKernels"
  • redundancy removed for model.setData(x,y)
  • rewrite "mean.proceed()" to "getMean()" and "getDerMatrix()"
  • rewrite "cov.proceed()" to "getCovMatrix()" and "getDerMatrix()"
  • rename cov.LIN to cov.Linear (to be consistent with mean.Linear)
  • rename module "valid" to "validation"
  • add graph dataset Mutag in python file. (.npz and .mat)
  • add graphUtil.nomalizeKernel()
  • fix number of iteration problem in graphKernels "PropagationKernel"
  • add unit testing for covariance, mean functions

bug fixes:

  • derivatives for cov.LINard
  • derivative of the scalar for cov.covScale
  • demo_GPR_FITC.py missing pyGPs.mean

July 8th 2014

structural updates:

  • add hyperparameter(signal variance s2) for linear covariance
  • add unit testing for inference,likelihood functions as well as models
  • NOT show(print) "maximum number of sweep warning in inference EP" any more
  • documentation updated

bug fixes:

  • typos in lik.Laplace
  • derivative in lik.Laplace

July 14th 2014

documentation updates:

  • online docs updated
  • API file updated

structural updates:

  • made private for methods that users don't need to call
July 17, 2014, 10:28:55
1.1

pyGPs v1.1 is released. It replaces pyGP_OO and contains substaintal updates in functionality and documentation. pyGP_PR v1.1 is released with substantial documentation updates and renamed (FN -> PR).

October 8, 2013, 12:35:28
1.0

Initial Announcement on mloss.org.

October 1, 2013, 14:12:14

Comments

No one has posted any comments yet. Perhaps you'd like to be the first?

Leave a comment

You must be logged in to post comments.