https://github.com/djsutherland/np-divs

Nonparametric nearest-neighbor-based estimators for divergences between distributions
https://github.com/djsutherland/np-divs

Last synced: 1 day ago
JSON representation

Nonparametric nearest-neighbor-based estimators for divergences between distributions

• Host: GitHub
• URL: https://github.com/djsutherland/np-divs
• Owner: djsutherland
• License: bsd-3-clause
• Created: 2012-02-03T03:12:48.000Z (almost 13 years ago)
• Default Branch: master
• Last Pushed: 2014-03-19T18:31:38.000Z (over 10 years ago)
• Last Synced: 2024-04-16T04:01:43.538Z (7 months ago)
• Language: C++
• Homepage:
• Size: 7.87 MB
• Stars: 4
• Watchers: 2
• Forks: 2
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: COPYING

Awesome Lists containing this project

README

        
This is a C++ implementation of the nonparametric divergence estimators

described by:

Barnabas Poczos, Liang Xiong, and Jeff Schneider (2011).

_Nonparametric divergence estimation with applications to machine learning on distributions._

Uncertainty in Artificial Intelligence.

http://autonlab.org/autonweb/20287.html

This code was written by Dougal J. Sutherland based on

[a pure-Matlab version](http://www.autonlab.org/autonweb/20466)

by the authors above.

Requirements

------------

  * [FLANN](http://people.cs.ubc.ca/~mariusm/index.php/FLANN/FLANN)

  * [Boost](http://boost.org) - at least 1.35

  * [CMake](http://cmake.org)

  * Optional: [HDF5](http://www.hdfgroup.org/HDF5/)

Installation

------------

    mkdir build; cd build

    cmake ..

    make

    make runtests # optional, requires HDF5

    make install

This will install the `npdivs` command-line interface (run `npdivs -h` for

help) as well as the shared library named e.g. `libnp-divs.so` (depending on

platform) and header files. By default, these will be placed in `/usr/local`;

to install to a different location, use something like:

    cmake .. -DCMAKE_INSTALL_PREFIX=$HOME

Note that when testing, the Gaussians50DTest case is disabled by default, as it

is computationally expensive and unlikely to reveal any installation problems

not shown by the Gaussians2DTest. If you'd like to run it anyway, use:

    make runtests GTEST_ALSO_RUN_DISABLED_TESTS=1