?? index.html
字號:
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN" "http://www.w3.org/TR/html4/strict.dtd"><html><head><link rev=made href="mailto:taku-ku@is.aist-nara.ac.jp"><title>TinySVM: Support Vector Machines</title><link rel=stylesheet href="tinysvm.css"></head><body><h1>TinySVM: Support Vector Machines</h1><p>Last update: $Date: 2002/08/20 06:14:22 $</p><hr><h2>TinySVM</h2><p>TinySVM is an implementation of Support Vector Machines (SVMs) <ahref="#vapnik95">[Vapnik 95]</a>, <a href="#vapnik98">[Vapnik 98]</a> for the problem of pattern recognition. Support Vector Machines is a new generation learning algorithmsbased on recent advances in statistical learning theory, and applied tolarge number of real-world applications, such as text categorization, hand-writtencharacter recognition.</p><h2>List of Contents</h2><ul><li><a href="#news">What's new</a><li><a href="#characteristics">Features</a><li><a href="libtinysvm.html">libtinysvm Reference Manual</a><li><a href="#usage-tools">Using included tools</a><li><a href="#download">Download</a> <ul> <li><a href="#source">Sources</a> <li><a href="#rpm">RPM package for RedHat Linux</a> <li><a href="#windows">Binary package for MS-Windows</a> <li><a href="#cvs">CVS</a> </ul><li><a href="#install">Install</a><li><a href="#bindings">Language Bindings</a> <ul> <li><a href="#perl">Perl</a> <li><a href="#ruby">Ruby</a> <li><a href="#python">Python</a> <li><a href="#java">Java</a> </ul><li><a href="#contacts">Questions and Bug Reports</a><li><a href="#todo">TODO</a><li><a href="#bib">References</a><li><a href="#links">Related Link</a></ul><h2><a name="news">What's new</a></h2><ul><li><strong>2002-08-20</strong>: <a href="#download">TinySVM 0.09</a> Released<br><ul> <li>Fix bug in -W option <li>Support Intel C++ compiler for Linux 6.0 (e.g. %env CC=icc CXX=icc ./configure --disable-shared; make ) <li>Support Borland C++ (use src/Makefile.bcc32) <li>Support Microsoft Visual Studio .NET/C++ (use src/Makefile.msvc) <li>Change extention of source files from .cc to .cpp</ul><li><strong>2002-03-08</strong><ul> <li>Support Mac OS X </ul><li><strong>2002-01-05</strong><ul> <li>Fix fatal bug in cache.h. (Thanks to Hideki Isozaki)</ul><li><strong>2001-12-07</strong><ul> <li>Support One-Class-SVM, (experimental) use -s option.</ul><li><strong>2001-09-03</strong><ul> <li>Support RBF, Neural, and ANOVA kernels <li>Script bindings for perl/ruby are rewritten with <a href="http://www.swig.org/">SWIG</a> <li>python and java interfaces are available</ul><li><strong>2001-08-25</strong>: 0.04<br><ul> <li>Fix memory leak bug in Claassify::classify(). <li>Implement a simple Garbage Collector (reference count) to handle training data effectively. <li>The following new options are added: <ul> <li>-I: svm_learn creates the optional file (MODEL.idx) which lists up alpha and G (gradient) of all training examples. Each line of MODEL.idx corresponds to the each training instance. <li>-M model_file: carry out incremental training with model_file and model_file.idx as a initial condition. You can reduce computational cost with this option.<pre>Sample:% svm_learn -I train model% lsmodel model.idx% cat new_train >> train% svm_learn -M model train model2</pre> <li>-W : When linear kernel is used, single vector w (w * x + b) is directly obtained instead of lists of alpha. </ul> <li>The following new API functions are added: <ul> <li>BaseExample::readSVindex() <li>BaseExample::writeSVindex() <li>BaseExample::set() <li>Example::rebuildSVindex() <li>Model::compress() </ul> <li>Add new API functions to perl/ruby interface, each of which corresponds to the original C++ new API.</ul><li><strong>2001-06-29</strong> 0.03<ul> <li>Delete -t option from svm_classify. <li>Bug fix in calculation of VC dimension.</ul><li><strong>2001-01-17</strong> 0.02<ul> <li>Support Support Vector Regression <li>Ruby module is available</ul></ul><h2><a name="characteristics">Features</a></h2><ul><li>Support standard C-SVR and C-SVM.<li>Uses sparse vector representation<li>Can handle several ten-thousands of training examples, and hundred-thousands of feature dimension.<li>Fast optimization algorithms stemming from <a href="#svmlight">SVM_light</a>,<a href="#joachims99">[Joachims/99a]</a>.<ul> <li>Working set selection based on steepest feasible descent. <li>"Shrinking" (an effective heuristics to reduce working sets dynamically)<a href="#joachims99">[Joachims/99a]</a> <li>Use LRU cache algorithm for store gram matrix.</ul><li>Optimization for handling binary features, twice faster than <a href="#svmlight">SVM_light</a>.<li>Provide well-known model selection criteria such as Leave-One-Out bound, VC dimension and Xi-Alpha estimation.<li>Written in C++ with OO style. Provides useful <a href="libtinysvm.html">class libraries</a>.<li>Provide Perl/Ruby/Python/Java module<li>Multi-platform, Unix/Windows</ul><h2><a name="usage-tools">Using the included Tools</a></h2><ul><h3>Format of training data</h3><p>TinySVM accepts the same representation of training data as <a href="#svmlight">SVM_light</a> uses.This format has potential to handle large sparse feature vectors.The format of training and test data file is:</p><p>(BNF-like representation)<br><class> .=. +1 | -1 <br><feature> .=. integer (>=1)<br><value> .=. real<br><line> .=. <class> <feature>:<value><feature>:<value> ... <feature>:<value><br></p><pre>Example (SVM)+1 201:1.2 3148:1.8 3983:1 4882:1-1 874:0.3 3652:1.1 3963:1 6179:1+1 1168:1.2 3318:1.2 3938:1.8 4481:1+1 350:1 3082:1.5 3965:1 6122:0.2-1 99:1 3057:1 3957:1 5838:0.3</pre>See tests/train.svmdata and tests/test.svmdata in the package.<p>In the case of SVR, you can give real value as class label</p><pre>Example (SVR)0.23 201:1.2 3148:1.8 3983:1 4882:10.33 874:0.3 3652:1.1 3963:1 6179:1-0.12 1168:1.2 3318:1.2 3938:1.8 4481:1</pre>See tests/train.svrdata in the package<h3><a name="svm_learn">svm_learn (learner)</a></h3><p>"svm_learn" accepts two arguments --- file name of training dataand model file in which the SVs and their weights (alpha) arestored after training. <br>Try --help option for finding out <a href="svm_learn.html">other options</a>.</p><pre>% svm_learn -t 1 -d 2 -c 1 train.svmdata modelTinySVM - tiny SVM packageCopyright (C) 2000 Taku Kudoh All rights reserved. 1000 examples, cache size: 1524.................... 1000 1000 0.0165 1286/ 64.3% 1286/ 64.3%............Checking optimality of inactive variables re-activated: 0Done! 1627 iterationsNumber of SVs (BSVs) 719 (4)Empirical Risk: 0.002 (2/1000)L1 Loss: 4.22975CPU Time: 0:00:01% ls -al model -rw-r--r-- 1 taku-ku is-stude 26455 Dec 7 13:40</pre>TinySVM prints the following messages during the learning iterations,<pre>.................... 1000 15865 2412 1.6001 33.2% 33.2%.................... 2000 15864 2412 1.3847 39.5% 36.4%</pre><ul><li>1st column: One <lit>"."</lit> means 50 iterations.<li>2nd column: Represents the total iterations processed<li>3rd column: Represents the size of current working set. It will become smaller during the <lit>shrinking</lit> process. <li>4th column: Represents the current cache size.<li>5th column: Max KKT violation value. If this value reaches<lit>termination-criterion</lit> (default 0.001), it means that the 1st stage ofthe learning process has completed.<li>7th column: Cache hit rate during last 1000 iterations. <li>7th column: Cache hit rate during all iterations.</ul><h3><a name="sary">svm_classify (simple classifier)</a></h3><p>"svm_classify" accepts two arguments --- file name of test data and model file generated by svm_learn."svm_classify" simply displays the accuracy of given test data. You can also employ interactive classification by giving "-" as file name of test example.<br>Try --help option for finding out <a href="svm_classify.html">other options</a>.</p><pre>% svm_classify test.svmdata modelAccuracy: 77.80000% (389/500)Precision: 66.82927% (137/205)Recall: 76.11111% (137/180)System/Answer p/p p/n n/p n/n: 137 68 43 252% svm_classify -V test.svmdata model-1 -1.04404-1 -1.26626-1 -0.545195.. snipAccuracy: 77.80000% (389/500)</pre><h3><a name="sary">svm_model (model browser)</a></h3><p>"svm_model" displays the estimated margin, VC dimension and number of SVsof given some model file. <br>Try --help option for finding out <a href="svm_model.html">other options</a>.</p><pre>% svm_model modelFile Name: modelMargin: 0.181666Number of SVs: 719Number of BSVs: 4Size of training data: 1000L1 Loss (Empirical Risk): 4.16917Estimated VC dimension: 728.219Estimated xi-alpha(2): 573</pre></ul><h2><a name="download">Download</a></h2><ul>TinySVM is free software distributed under the <a href="http://www.gnu.org/copyleft/lesser.html">GNU Lesser General Public License</a>.<h3><a name="source">Source</a></h3><ul><li>TinySVM-0.09.tar.gz:<a href="./src/TinySVM-0.09.tar.gz">HTTP</a></ul><h3><a name="rpm">Binary/Source package for RedHat Linux</a></h3><ul><li>RedHat 6.x i386:<a href="./redhat-6.x/RPMS/i386/">HTTP</a><li>RedHat 6.x SRPMS: <a href="./redhat-6.x/SRPMS/">HTTP</a><li>RedHat 7.x i386:<a href="./redhat-7.x/RPMS/i386/">HTTP</a><li>RedHat 7.x SRPMS: <a href="./redhat-7.x/SRPMS/">HTTP</a></ul><h3><a name="rpm">Binary package for MS-Windows</a></h3><ul><li><a href="./win/">HTTP</a></ul><h3><a name="cvs">CVS</a></h3><ul>Development of TinySVM uses CVS. So latest developing version isavailable at CVS. <br>We are willing to welcome you to join CVS based development.<pre>% cvs -d :pserver:anonymous@chasen.aist-nara.ac.jp:/cvsroot loginCVS password: # Just hit return/enter.% cvs -d :pserver:anonymous@chasen.aist-nara.ac.jp:/cvsroot co TinySVM</pre></ul></ul><h2><a name="install">Install</a></h2><ul><pre> % ./configure % make % make check % su # make install</pre>You can change default install path by using --prefix option ofconfigure script. <br>Try --help option for finding out other options.</ul><h2><a name="bindings">Language Bindings</a></h2><ul><li><a name="perl">Perl</a>: see perl/README file.<li><a name="ruby">Ruby</a>: see ruby/README file<li><a name="python">Python</a>: see python/README file<li><a name="java">Java</a>: see java/README file</ul><h2><a name="contacts">Questions and Bug Reports</a></h2><ul>If you find bugs or you have any questionsplease contact me via email taku-ku@is.aist-nara.ac.jp.<br>(Japanese mail is also (more) acceptable.)</ul><h2><a name="todo">TODO</a></h2><ul><li>MultiClass SVM (one-for-all-others, pairwise)<li>nu-SVM and nu-SVR <a href="#scholkopf1998">[Schölkopf1998]</a><li>Transductive SVM <a href="#vapnik98">[Vapnik 98]</a>,<a href="#joachims99c">[Joachims 99]</a><li>Span SVM <a href="#vapnik2000">[Vapnik 2000]</a><li>Ordinal SVR <a href="#herbrich2000">[Herbrich 2000]</a><li>Provide some Wrapper class to handle STRING features.<li>Provide DLL (Dynamic Link Libraries) for Windows environment.<li>Provide an API for user customizable kernel function.<br></ul><h2><a name="bib">References</a></h2><ul><li> <a name="joachims99">[Joachims 99a] T. Joachims, 11 in: Making large-Scale SVM Learning Practical. Advances in Kernel Methods pp.169-</a><li> <a name="vapnik95">[Vapnik 95] Vladimir N. Vapnik, The Nature of Statistical Learning Theory. Springer, 1998.</a><li> <a name="vapnik95">[Vapnik 98] Vladimir N. Vapnik, The Statisitcal Learning Theory. Springer, 1998.</a><li> <a name="joachims99c">[Joachims 99c] T. Joachims, Transductive Inference for Text Classification using Support Vector Machines. International Conference on Machine Learning (ICML), 1999.</a><li> <a name="vapnik2000">[Vapnik 2000] Vladimir N. Vapnik, 14 in: Bounds on Error Expectation for SVM. in Advances in Large Margin Classifiers. pp. 261-</a><li> <a name="herbrich2000">[Herbrich 2000] Ralf Herbrich et al, 7 in: Large Margin Rank Boundaries for Ordinal Regression. in Advances in Large Margin Classifiers. pp. 116-</a><li> <a name="scholkopf1998">[Schölkopf 1998] B. Schölkopf, A. Smola, R. Williamson, and P. L. Bartlett. New support vector algorithms. NeuroCOLT Technical Report NC-TR-98-031, Royal Holloway College, University of London, UK, 1998. To appear in Neural Computation.</a></ul><h2><a name="bib">Related Links</a></h2><ul><li> <a name="svmlight"><a href="http://www-ai.cs.uni-dortmund.de/SOFTWARE/SVM_LIGHT/index.eng.html">SVM_light</a></a><li> <a href="http://www.csie.ntu.edu.tw/~cjlin/libsvm">libsvm</a> Yet another Fast and useful SVM package<li><a href="http://www.kernel-machines.org/">Kernel Machine Website</a><li><a href="http://www.clopinet.com/isabelle/Projects/SVM/applist.html">SVM Application List</a><li><a href="http://svm.research.bell-labs.com/">Support Vectors and Kernel Methods</a><li><a href="http://www.support-vector.net/">An introduction to Suppor Vector Machines</a></ul><hr><p>$Id: index.html,v 1.26 2002/08/20 06:14:22 taku-ku Exp $;</p><address>taku-ku@is.aist-nara.ac.jp</address></body></html>?? 快捷鍵說明
復制代碼
Ctrl + C
搜索代碼
Ctrl + F
全屏模式
F11
切換主題
Ctrl + Shift + D
顯示快捷鍵
?
增大字號
Ctrl + =
減小字號
Ctrl + -