https://github.com/patwie/robustkernelpaths

ICML paper: Robust and Efficient Kernel Hyperparameter Paths with Guarantees
https://github.com/patwie/robustkernelpaths

eigen hyperparameters icml kernel kernel-hyperparameter libsvm machine-learning machine-learning-algorithms

Last synced: 16 days ago
JSON representation

ICML paper: Robust and Efficient Kernel Hyperparameter Paths with Guarantees

• Host: GitHub
• URL: https://github.com/patwie/robustkernelpaths
• Owner: PatWie
• License: lgpl-3.0
• Created: 2014-08-18T18:41:47.000Z (about 10 years ago)
• Default Branch: master
• Last Pushed: 2023-01-10T09:21:26.000Z (almost 2 years ago)
• Last Synced: 2024-10-08T21:42:29.724Z (about 1 month ago)
• Topics: eigen, hyperparameters, icml, kernel, kernel-hyperparameter, libsvm, machine-learning, machine-learning-algorithms
• Language: C++
• Homepage: http://patwie.com
• Size: 31.3 KB
• Stars: 2
• Watchers: 3
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Robust and Efficient Kernel Hyperparameter Paths with Guarantees

**auto finetuning, svm, parameters, kernel parameter, choose parameters **

The results of non-linear SVM classifications heavily rely on the choose of the kernel hyperparameter, i.e. , \lambda in $k(x,y)=\exp(-\lambda \Vert{x-y}\Vert²)$ in the kernel function.

This repository contains an effective algorithmn that calculates an approximate entire solution path of the objective function with respect to the hyperparameter within the interval $[2^{-10},2^{10}]$ without numerical issues by which the exact algorithms suffer.

More details can be found in the corresponding [paper][paper]. 

For the matrix calculation the [library Eigen][eigen] was used in combination with [OpenMP][openmp]. The backend solver is [libSVM][libsvm]

The program reads the problem description in the default libSVM format. See the documented source for more information or run the compiled program without any parameters to get help.

### weblinks

 * [jmlr](http://jmlr.org/proceedings/papers/v32/giesen14.html) entry

 * [paper](http://jmlr.org/proceedings/papers/v32/giesen14.pdf) pdf format

 

[jmlr]:http://jmlr.org/proceedings/papers/v32/giesen14.html

[paper]:http://jmlr.org/proceedings/papers/v32/giesen14.pdf

[eigen]:http://eigen.tuxfamily.org/index.php?title=Main_Page

[libsvm]:http://www.csie.ntu.edu.tw/~cjlin/libsvm/

[openmp]:http://openmp.org/wp/