Skip to content

The Mondrian kernel is a random feature approximation to the Laplace kernel allowing fast kernel width selection.

License

Notifications You must be signed in to change notification settings

matejbalog/mondrian-kernel

Repository files navigation

The Mondrian Kernel

Matej Balog, Balaji Lakshminarayanan, Zoubin Ghahramani, Daniel M. Roy, Yee Whye Teh

Thirty-Second Conference on Uncertainty in Artificial Intelligence (UAI), 2016.

[PDF] [supp] [arXiv] [poster] [slides]

The scripts provided here implement experiments from this paper. The scripts experiment_1_laplace_kernel_approximation, experiment_2_fast_kernel_width_learning and experiment_3_mondrian_kernel_vs_forest are intended to be directly runnable.

Requirements

Python packages: heapq, matplotlib, numpy, scipy, sklearn, sys, time

The CPU dataset cpu.mat can be download and extracted from here.

Known issues

A bug in scipy may cause the Python kernel to restart when loading the CPU dataset from cpu.mat. Downgrading to scipy 0.16.0 should solve the problem.

BibTeX

@inproceedings{balog2016mondriankernel,
  author = {Matej Balog and Balaji Lakshminarayanan and Zoubin Ghahramani and Daniel M.~Roy and Yee Whye Teh},
  title={The {M}ondrian Kernel},
  booktitle = {32nd Conference on Uncertainty in Artificial Intelligence (UAI)},
  year = {2016},
  month = {June},
  url = {http://www.auai.org/uai2016/proceedings/papers/236.pdf}
}

About

The Mondrian kernel is a random feature approximation to the Laplace kernel allowing fast kernel width selection.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages