proportion-SVM (pSVM, $\propto$SVM) for learning with label proportions
Matlab M
Switch branches/tags
Nothing to show
Latest commit 9092eec Jul 12, 2013 @felixyu added readme file
Permalink
Failed to load latest commit information.
lib initial commit Jul 11, 2013
alternating_svm_dual_with_slack.m initial commit Jul 11, 2013
demo_heart.m initial commit Jul 11, 2013
demo_toy.m initial commit Jul 11, 2013
heart.mat initial commit Jul 11, 2013
init.m initial commit Jul 11, 2013
invcal_dual.m initial commit Jul 11, 2013
invcal_dual_predict.m initial commit Jul 11, 2013
readme.md added readme file Jul 11, 2013
test_all_method.m initial commit Jul 11, 2013
toy.mat initial commit Jul 11, 2013

readme.md

In the interest of time, I am releasing a preliminary version of pSVM (felixyu.org/pSVM.html).

Please cite the following paper: Felix X. Yu; Dong Liu; Sanjiv Kumar; Tony Jebara; Shih-Fu Chang. $\propto$SVM for learning with label proportions, ICML 2013

  1. The package implements the (slow) kernel version of InvCal, alter-pSVM and conv-pSVM. Linear versions, faster kernel versions based on libsvm (instead of gurobi) will be released later.

  2. Setup The code requires CVX, gurobi, and libsvm. http://cvxr.com/cvx/ http://www.gurobi.com/ http://www.csie.ntu.edu.tw/~cjlin/libsvm/ Please download the software in the above websites, and setup init.m accordingly.

  3. Demo Run demo_toy and demo_heart for two simple demos. The toy data was the one used in the ICML paper.

  4. The conv-pSVM was modified from LGMMC(http://lamda.nju.edu.cn/code_LGMMC.ashx). (Y.-F. Li, I. W. Tsang, J. T. Kwok, and Z.-H. Zhou. Tighter and convex maximum margin clustering. AISTATS'09) Some of their code is included for the ease of the users. All the terms of the LGMMC package apply.

  5. Please direct any questions to Felix Yu (yuxinnan@ee.columbia.edu).