<!doctype html public "-//w3c//dtd html 4.0 transitional//en"> <html> <head> <meta http-equiv="Content-Type" content="text/html; charset=iso-8859-1"> <meta name="description" content="logistic regression, support vector machines, linear classification, document classification"> <title>LIBLINEAR Experiments </title> </head> <body text="#000000" bgcolor="#FFEFD5" link="#FF0000" vlink="#0000FF"> <h2> LIBLINEAR Experiments </h2> <h3> <a href="http://www.csie.ntu.edu.tw/~cjlin/mlgroup">Machine Learning Group</a> at National Taiwan University <br> </h3> <p> This page provides the source codes for the papers related to <b>LIBLINEAR</b>. <hr> <h3> <a name=linear_ranksvm> Experiments on linear rankSVM</h3> Programs used to generate experiment results in the paper C.-P. Lee and C.-J. Lin. <a href=../papers/ranksvm/ranksvml2.pdf> Large-scale Linear RankSVM</a>. Technical report, 2013. <p> can be found in this <a href=../papers/ranksvm/ranksvml2_exp-1.3.tgz>tar.gz file</a>. <p> Use files here only if you are interested in redoing our experiments. To apply the method for your applications, all you need is a LIBLINEAR extension. Check "Large-scale linear rankSVM" at <a href=../libsvmtools>LIBSVM Tools</a>. <hr> <h3> <a name=linear_svr> Experiments on linear support vector regression</h3> Programs used to generate experiment results in the paper C.-H. Ho, and C.-J. Lin. <a href=../papers/linear-svr.pdf> Large-scale Linear Support Vector Regression</a>. JMLR, 2012. <p> can be found in this <a href="http://www.csie.ntu.edu.tw/~cjlin/liblinear/exps/svr/linear_svr_exp-1.2.zip">zip file</a>. <hr> <h3> <a name=disk_decomposition> Experiments on linear classification when data cannot fit in memory</h3> An algorithm in <p> H.-F. Yu, C.-J. Hsieh, K.-W. Chang, and C.-J. Lin</b>, <A HREF="../papers/disk_decomposition/tkdd_disk_decomposition.pdf"> Large linear classification when data cannot fit in memory</a>. ACM KDD 2010 (Best research paper award). Extended version appeared in <a href=http://portal.acm.org/tkdd/>ACM Transactions on Knowledge Discovery from Data</a>, 5:23:1--23:23, 2012. <p> has been implemented as an extension of LIBLINEAR. It aims to handle data larger than your memory capacity. It can be found in <a href=../libsvmtools>LIBSVM Tools</a>. <p> To repeat experiments in our paper, check this <a href="http://www.csie.ntu.edu.tw/~cjlin/liblinear/exps/cdblock/cdblock_exp-2.0.tgz">tgz file</a>. Don't use it unlesse you want to regenerate figures. For you own experiments, you should use the LIBLINEAR extension at LIBSVM tools. <hr> <h3> <a name=maxent_dual_exp> Experiments on Dual Logistic Regression and Maximum Entropy</h3> Programs used to generate experiment results in the paper <p> Hsiang-Fu Yu, Fang-Lan Huang, and Chih-Jen Lin. <a href=../papers/maxent_dual.pdf> Dual Coordinate Descent Methods for Logistic Regression and Maximum Entropy Models </a>. <I><A HREF= "http://www.springer.com/computer/ai/journal/10994"> Machine Learning</A></I>, 85(2011), 41-75. <p> can be found in this <a href="http://www.csie.ntu.edu.tw/~cjlin/liblinear/maxent/maxent_dual_exp-1.0.zip">zip file</a>. <hr> <h3> <a name=l1_exp> Comparing Large-scale L1-regularized Linear Classifiers </h3> <ul> <li> The following paper compares various L1-regularized solvers for logistic regression and SVM. The algorithm CDN used in <b>LIBLINEAR</b> now for L1-regularized SVM was proposed here. <p> Guo-Xun Yuan, Kai-Wei Chang, Cho-Jui Hsieh, and Chih-Jen Lin. <a href=../papers/l1.pdf> A Comparison of Optimization Methods for Large-scale L1-regularized Linear Classification.</a> JMLR 2010. <p> Programs for generating experimental results can be found in this <a href="http://www.csie.ntu.edu.tw/~cjlin/liblinear/l1paper/l1_exp-1.2.zip">zip file</a>. </li> <li> For L1-regularized logistic regression, the following paper proposes an algorithm (newGLMNET) to succeed CDN in <b>LIBLINEAR</b>. <p> Guo-Xun Yuan, Chia-Hua Ho, and Chih-Jen Lin. <a href=../papers/long-glmnet.pdf> An Improved GLMNET for L1-regularized Logistic Regression and Support Vector Machines. </a> JMLR, 2012 <p> Programs for generating experimental results can be found in this <a href="http://www.csie.ntu.edu.tw/~cjlin/liblinear/l1paper/newGLMNET_exp-1.0.zip">zip file</a>. </li> </ul> <p> You can directly use LIBLINEAR for efficient L1-regularized classification. Use code here <b>only</b> if you are interested in redoing our experiments. The running time is long because we run each solver to accurately solve optimization problems. <hr> <h3> <a name=lowpoly_exp> Experiments on Degree-2 Polynomial Mappings of Data</h3> Programs used to generate experiment results in Section 5 of the paper <p> Yin-Wen Chang, Cho-Jui Hsieh, Kai-Wei Chang, Michael Ringgaard and Chih-Jen Lin. <a href=../papers/lowpoly_journal.pdf> Low-Degree Polynomial Mapping of Data for SVM</a>, JMLR 2010, <p> can be found in this <a href="http://www.csie.ntu.edu.tw/~cjlin/liblinear/lowpoly/lowpoly_exp-1.1.zip">zip file</a>. <p> Use files here only if you are interested in redoing our experiments. To apply the method for your applications, all you need is a LIBLINEAR extension. Check "fast training/testing of degree-2 polynomial mappings of data" at <a href=../libsvmtools>LIBSVM Tools</a>. <hr> <h3> <a name=maxent_exp> Experiments on Maximum Entropy models</h3> Programs used to generate experiment results in the paper <p> Fang-Lan Huang, Cho-Jui Hsieh, Kai-Wei Chang, and Chih-Jen Lin. <a href=../papers/maxent_journal.pdf> Iterative Scaling and Coordinate Descent Methods for Maximum Entropy Models</a>, JMLR 2010, <p> can be found in this <a href="http://www.csie.ntu.edu.tw/~cjlin/liblinear/maxent/maxent_exp-1.0.zip">zip file</a>. <hr> <h3> <a name=dual_exp> Comparing various methods for large-scale linear SVM</h3> Programs used to generate experiment results in the paper <p> C.-J. Hsieh, K.-W. Chang, C.-J. Lin, S. Sundararajan, and S. Sathiya Keerthi. <a href=../papers/cddual.pdf> A Dual Coordinate Descent Method for Large-scale Linear SVM</a>, ICML 2008, <p> can be found in this <a href="http://www.csie.ntu.edu.tw/~cjlin/liblinear/dualpaper/dual_exp-1.0.zip">zip file</a>. <hr> <h3> <a name=cdl2_exp> Comparing various methods for large-scale linear SVM</h3> Programs used to generate experiment results in the paper <p> K.-W. Chang, C.-J. Hsieh, and C.-J. Lin, <a href=../papers/cdl2.pdf> Coordinate Descent Method for Large-scale L2-loss Linear SVM </a>, JMLR 2008, <p> can be found in this <a href="liblinear/cdl2paper">zip file</a>. <hr> <h3> <a name=lrpaper> Comparing various methods for logistic regression</h3> Programs used to generate experiment results in the paper <p> C.-J. Lin, R. C. Weng, and S. S. Keerthi. <a href=../papers/logistic.pdf> Trust region Newton method for large-scale logistic regression</a>, JMLR 2008, <p> can be found in this <a href="http://www.csie.ntu.edu.tw/~cjlin/liblinear/lrpaper/lrpaper-1.04.zip">zip file</a>. <p> We include <a href="http://www.ece.northwestern.edu/~nocedal/lbfgs.html">LBFGS</a> and <a href="http://people.cs.uchicago.edu/~vikass/svmlin.html">SVMlin</a> (a <b>modified</b> version) for experiments. Please check their respective COPYRIGHT notices. <hr> Please send comments and suggestions to <a href="../index.html">Chih-Jen Lin</a>. </body> </html>