Libsvm 进行 SVM 分类.rar_分类_其他

<html xmlns="http://www.w3.org/1999/xhtml"> <head> <meta charset="utf-8"> <meta name="generator" content="pdf2htmlEX"> <meta http-equiv="X-UA-Compatible" content="IE=edge,chrome=1"> <link rel="stylesheet" href="https://static.pudn.com/base/css/base.min.css"> <link rel="stylesheet" href="https://static.pudn.com/base/css/fancy.min.css"> <link rel="stylesheet" href="https://static.pudn.com/prod/directory_preview_static/6250fdb26caf5961923369ce/raw.css"> <script src="https://static.pudn.com/base/js/compatibility.min.js"></script> <script src="https://static.pudn.com/base/js/pdf2htmlEX.min.js"></script> <script> try{ pdf2htmlEX.defaultViewer = new pdf2htmlEX.Viewer({}); }catch(e){} </script> <title></title> </head> <body> <div id="sidebar" style="display: none"> <div id="outline"> </div> </div> <div id="pf1" class="pf w0 h0" data-page-no="1"><div class="pc pc1 w0 h0"><img class="bi x0 y0 w1 h1" alt="" src="https://static.pudn.com/prod/directory_preview_static/6250fdb26caf5961923369ce/bg1.jpg"><div class="t m0 x1 h2 y1 ff1 fs0 fc0 sc0 ls0 ws0">www.theme<span class="_ _0"></span>gall<span class="_ _0"></span>ery.com</div><div class="t m0 x2 h3 y2 ff2 fs1 fc1 sc0 ls0 ws0">LOGO</div><div class="t m0 x3 h4 y3 ff3 fs2 fc2 sc1 ls0 ws0">教你认识<span class="ff4 sc0">SVM<span class="_ _0"></span><span class="ff3 sc1">和使<span class="_ _0"></span>用开源工<span class="_ _0"></span>具<span class="ff4 sc0">LibSVM</span></span></span></div><div class="t m0 x4 h5 y4 ff5 fs1 fc1 sc0 ls0 ws0">夏睿<span class="_"> </span><span class="ff6">rxia@nlpr.ia.ac.cn</span></div></div><div class="pi" data-data='{"ctm":[1.140576,0.000000,0.000000,1.140576,0.000000,0.000000]}'></div></div> </body> </html>

<div id="pf2" class="pf w0 h0" data-page-no="2"><div class="pc pc2 w0 h0"><img class="bi x0 y0 w1 h1" alt="" src="https://static.pudn.com/prod/directory_preview_static/6250fdb26caf5961923369ce/bg2.jpg"><div class="t m0 x5 h6 y5 ff7 fs0 fc0 sc0 ls0 ws0">www.them<span class="_ _0"></span>egal<span class="_ _0"></span>lery<span class="_ _0"></span>.com</div><div class="t m0 x6 h7 y6 ff8 fs3 fc2 sc0 ls0 ws0">v<span class="_ _1"> </span><span class="ff6 fc0">SVM<span class="ff5">能<span class="_ _2"></span>解决什么<span class="_ _2"></span>问题？</span></span></div><div class="t m0 x7 h7 y7 ff8 fs3 fc3 sc0 ls0 ws0">§<span class="_ _3"> </span><span class="ff5 fc0">从机器学<span class="_ _2"></span>习谈起</span></div><div class="t m0 x7 h7 y8 ff8 fs3 fc3 sc0 ls0 ws0">§<span class="_ _3"> </span><span class="ff5 fc0">分类与回<span class="_ _2"></span>归</span></div><div class="t m0 x6 h7 y9 ff8 fs3 fc2 sc0 ls0 ws0">v<span class="_ _1"> </span><span class="ff5 fc0">为什么选<span class="_ _2"></span>择<span class="_ _2"></span><span class="ff6">SVM</span>？</span></div><div class="t m0 x7 h7 ya ff8 fs3 fc3 sc0 ls0 ws0">§<span class="_ _3"> </span><span class="ff5 fc0">从<span class="ff6">AN<span class="_ _2"></span>N</span>到<span class="_ _2"></span><span class="ff6">SVM</span></span></div><div class="t m0 x7 h7 yb ff8 fs3 fc3 sc0 ls0 ws0">§<span class="_ _3"> </span><span class="ff6 fc0">SVM<span class="ff5">的<span class="_ _2"></span>两个核心<span class="_ _2"></span>思想</span></span></div><div class="t m0 x7 h7 yc ff8 fs3 fc3 sc0 ls0 ws0">§<span class="_ _3"> </span><span class="ff6 fc0">SVM<span class="ff5">算<span class="_ _2"></span>法</span></span></div><div class="t m0 x6 h7 yd ff8 fs3 fc2 sc0 ls0 ws0">v<span class="_ _1"> </span><span class="ff5 fc0">怎么用<span class="_ _2"></span><span class="ff6">SVM</span>？</span></div><div class="t m0 x7 h7 ye ff8 fs3 fc3 sc0 ls0 ws0">§<span class="_ _3"> </span><span class="ff5 fc0">用<span class="ff6">LibSVM</span>做<span class="_ _2"></span>分<span class="_ _2"></span>类</span></div><div class="t m0 x7 h7 yf ff8 fs3 fc3 sc0 ls0 ws0">§<span class="_ _3"> </span><span class="ff5 fc0">用<span class="ff6">LibSVM</span>做<span class="_ _2"></span>回<span class="_ _2"></span>归</span></div><div class="t m0 x7 h7 y10 ff8 fs3 fc3 sc0 ls0 ws0">§<span class="_ _3"> </span><span class="ff6 fc0">Python<span class="ff5">平台<span class="_ _2"></span>下的<span class="_ _2"></span></span>LibSVM</span></div><div class="t m0 x8 h8 y11 ff3 fs2 fc1 sc2 ls0 ws0">提纲</div></div><div class="pi" data-data='{"ctm":[1.140576,0.000000,0.000000,1.140576,0.000000,0.000000]}'></div></div>

<div id="pf3" class="pf w0 h0" data-page-no="3"><div class="pc pc3 w0 h0"><img class="bi x0 y0 w1 h1" alt="" src="https://static.pudn.com/prod/directory_preview_static/6250fdb26caf5961923369ce/bg3.jpg"><div class="t m0 x5 h6 y5 ff7 fs0 fc0 sc0 ls0 ws0">www.them<span class="_ _0"></span>egal<span class="_ _0"></span>lery<span class="_ _0"></span>.com</div><div class="t m0 x8 h8 y11 ff3 fs2 fc1 sc2 ls0 ws0">从机器学习谈起</div><div class="t m0 x9 h9 y12 ff8 fs4 fc2 sc0 ls0 ws0">v<span class="_ _4"> </span><span class="ff5 fc0">目的</span></div><div class="t m0 xa ha y13 ff5 fs4 fc0 sc0 ls0 ws0">根据给定的训练样本，对某系统输入输出之间依赖关系的估计，使它能</div><div class="t m0 xa ha y14 ff5 fs4 fc0 sc0 ls0 ws0">够对未知输出作出尽可能准确的预测。</div><div class="t m0 x9 h9 y15 ff8 fs4 fc2 sc0 ls0 ws0">v<span class="_ _4"> </span><span class="ff5 fc0">机器学习的三个基本问题</span></div><div class="t m0 xb h9 y16 ff8 fs4 fc3 sc0 ls0 ws0">§<span class="_ _5"> </span><span class="ff5 fc0">模式识别<span class="ff6">——</span>分类</span></div><div class="t m0 xb h9 y17 ff8 fs4 fc3 sc0 ls0 ws0">§<span class="_ _5"> </span><span class="ff5 fc0">函数拟合<span class="ff6">——</span>回归</span></div><div class="t m0 xb h9 y18 ff8 fs4 fc3 sc0 ls0 ws0">§<span class="_ _5"> </span><span class="ff5 fc0">概率密度估计</span></div><div class="t m0 x9 h9 y19 ff8 fs4 fc2 sc0 ls0 ws0">v<span class="_ _4"> </span><span class="ff5 fc0">数学表述</span></div><div class="t m0 xb h9 y1a ff8 fs4 fc3 sc0 ls0 ws0">§<span class="_ _5"> </span><span class="ff5 fc0">给定条件：<span class="ff6">n</span>个独立同分布观测样本</span></div><div class="t m0 xb h9 y1b ff8 fs4 fc3 sc0 ls0 ws0">§<span class="_ _5"> </span><span class="ff5 fc0">目标：求一个最优函数</span></div><div class="t m0 xb h9 y1c ff8 fs4 fc3 sc0 ls0 ws0">§<span class="_ _5"> </span><span class="ff5 fc0">最理想的要求：最小化期望风险</span></div><div class="t m0 x9 h9 y1d ff8 fs4 fc2 sc0 ls0 ws0">v<span class="_ _4"> </span><span class="ff5 fc0">机器学习方法的代表</span></div><div class="t m0 xb h9 y1e ff8 fs4 fc3 sc0 ls0 ws0">§<span class="_ _5"> </span><span class="ff5 fc0">人工神经网络（<span class="ff6">ANN</span>）<span class="ff6">& </span>支持向量机（<span class="_ _2"></span><span class="ff6">SVM</span>）</span></div><div class="t m0 xc hb y1f ff6 fs5 fc0 sc0 ls0 ws0">1<span class="_ _6"> </span>1<span class="_ _7"> </span>2<span class="_ _8"> </span>2</div><div class="c xd y20 w2 hc"><div class="t m0 x0 hd y21 ff6 fs6 fc0 sc0 ls0 ws0">(<span class="_ _9"> </span>,<span class="_ _a"> </span>),<span class="_ _b"></span>(<span class="_ _c"> </span>,<span class="_ _d"> </span>),<span class="_ _e"> </span>(<span class="_ _c"> </span>,<span class="_ _e"> </span>)</div></div><div class="c xe y22 w3 he"><div class="t m0 x0 hf y23 ff9 fs5 fc0 sc0 ls0 ws0">n<span class="_ _8"> </span>n</div></div><div class="c xf y20 w4 hc"><div class="t m0 x0 h10 y21 ff9 fs6 fc0 sc0 ls0 ws0">x<span class="_ _3"> </span>y<span class="_ _f"> </span>x<span class="_ _10"> </span>y<span class="_ _11"> </span>x<span class="_ _10"> </span>y</div></div><div class="c x10 y24 w5 h11"><div class="t m0 x0 h12 y25 ffa fs6 fc0 sc0 ls0 ws0">L</div></div><div class="c x11 y26 w6 he"><div class="t m0 x0 h13 y27 ff6 fs7 fc0 sc0 ls0 ws0">*</div></div><div class="c x12 y28 w7 h14"><div class="t m0 x0 h15 y29 ff6 fs8 fc0 sc0 ls0 ws0">(<span class="_ _12"> </span>,<span class="_ _13"> </span>)</div></div><div class="c x13 y28 w8 h14"><div class="t m0 x14 h16 y29 ff9 fs8 fc0 sc0 ls0 ws0">f<span class="_ _14"> </span>x<span class="_ _15"> </span>w</div></div><div class="c x15 y2a w9 h17"><div class="t m0 x0 h18 y2b ff6 fs9 fc0 sc0 ls0 ws0">(<span class="_ _4"> </span>)<span class="_ _16"> </span>(<span class="_ _1"> </span>,<span class="_ _17"> </span>(<span class="_ _18"> </span>,<span class="_ _19"> </span>))<span class="_ _1a"> </span>(<span class="_ _18"> </span>,<span class="_ _4"> </span>)</div></div><div class="c x16 y2a wa h17"><div class="t m0 x0 h19 y2b ff9 fs9 fc0 sc0 ls0 ws0">R<span class="_ _1b"> </span>w<span class="_ _1c"> </span>L<span class="_ _1d"> </span>y<span class="_ _1"> </span>f<span class="_ _1e"> </span>x<span class="_ _1b"> </span>w<span class="_ _1f"> </span>dF<span class="_ _20"> </span>x<span class="_ _1d"> </span>y</div></div><div class="t m0 x17 h1a y2c ffb fs9 fc0 sc0 ls0 ws0">=</div><div class="c x18 y2d wb h1b"><div class="t m0 x0 h1c y2e ffb fsa fc0 sc0 ls0 ws0">∫</div></div></div><div class="pi" data-data='{"ctm":[1.140576,0.000000,0.000000,1.140576,0.000000,0.000000]}'></div></div>

<div id="pf4" class="pf w0 h0" data-page-no="4"><div class="pc pc4 w0 h0"><img class="bi x0 y0 w1 h1" alt="" src="https://static.pudn.com/prod/directory_preview_static/6250fdb26caf5961923369ce/bg4.jpg"><div class="t m0 x5 h6 y5 ff7 fs0 fc0 sc0 ls0 ws0">www.them<span class="_ _0"></span>egal<span class="_ _0"></span>lery<span class="_ _0"></span>.com</div><div class="t m0 x8 h4 y2f ff4 fs2 fc1 sc0 ls0 ws0">ANN VS S<span class="_ _0"></span>VM</div><div class="t m0 x19 h1d y30 ff8 fsb fc0 sc0 ls0 ws0">à</div><div class="t m0 x1a h1e y31 ff5 fsb fc0 sc0 ls0 ws0">全局最优解</div><div class="t m0 x19 h1d y32 ff8 fsb fc0 sc0 ls0 ws0">à</div><div class="t m0 x1b h1e y31 ff5 fsb fc0 sc0 ls0 ws0">容易陷入局部最<span class="_ _2"></span>小</div><div class="t m0 x1c h1e y33 ff5 fsb fc0 sc0 ls0 ws0">训练较快</div><div class="t m0 x19 h1d y34 ff8 fsb fc0 sc0 ls0 ws0">à</div><div class="t m0 x1d h1e y33 ff5 fsb fc0 sc0 ls0 ws0">训练较慢</div><div class="t m0 x19 h1d y35 ff8 fsb fc0 sc0 ls0 ws0">à</div><div class="t m0 x19 h1d y36 ff8 fsb fc0 sc0 ls0 ws0">à</div><div class="t m0 x19 h1d y37 ff8 fsb fc0 sc0 ls0 ws0">à</div><div class="t m0 x19 h1d y38 ff8 fsb fc0 sc0 ls0 ws0">à</div><div class="t m0 x1e h1f y39 ff6 fsb fc0 sc0 ls0 ws0">VS</div><div class="t m0 x1f h1e y3a ff5 fsb fc0 sc0 ls0 ws0">这个黑匣子要透<span class="_ _2"></span>明一点<span class="_ _2"></span>，怎么<span class="_ _2"></span>个透</div><div class="t m0 x20 h1f y3b ff5 fsb fc0 sc0 ls0 ws0">明法？<span class="_ _21"> </span><span class="fc4">（过人之处<span class="ff6">2<span class="_ _2"></span></span>）</span></div><div class="t m0 x21 h1e y3a ff5 fsb fc0 sc0 ls0 ws0">完全的黑匣子，<span class="_ _2"></span>高度非<span class="_ _2"></span>线性，<span class="_ _2"></span>人工难</div><div class="t m0 x22 h1e y3c ff5 fsb fc0 sc0 ls0 ws0">以理解和干预</div><div class="t m0 x23 h1f y36 ff5 fsb fc0 sc0 ls0 ws0">推广能力的控制<span class="_ _21"> </span><span class="fc4">（<span class="_ _2"></span>过人之<span class="_ _2"></span>处<span class="ff6">1</span>）</span></div><div class="t m0 x24 h1e y3d ff5 fsb fc0 sc0 ls0 ws0">过学习</div><div class="t m0 x20 h1f y3e ff5 fsb fc0 sc0 ls0 ws0">结构风险最小化<span class="_ _2"></span><span class="ff6">(SRM)<span class="_ _22"></span><span class="ff5">经验风险最小化<span class="_ _2"></span><span class="ff6">(ERM)</span></span></span></div><div class="t m0 x25 h1e y3f ff5 fsb fc0 sc0 ls0 ws0">统计学习理论<span class="_ _23"></span>传统统计学</div><div class="t m0 x26 h1f y40 ff6 fsb fc0 sc0 ls0 ws0">SVM<span class="_ _24"></span>ANN</div></div><div class="pi" data-data='{"ctm":[1.140576,0.000000,0.000000,1.140576,0.000000,0.000000]}'></div></div>

<div id="pf5" class="pf w0 h0" data-page-no="5"><div class="pc pc5 w0 h0"><img class="bi x0 y0 w1 h1" alt="" src="https://static.pudn.com/prod/directory_preview_static/6250fdb26caf5961923369ce/bg5.jpg"><div class="t m0 x5 h6 y5 ff7 fs0 fc0 sc0 ls0 ws0">www.them<span class="_ _0"></span>egal<span class="_ _0"></span>lery<span class="_ _0"></span>.com</div><div class="t m0 x8 h8 y11 ff3 fs2 fc1 sc2 ls0 ws0">广义最优分类面</div><div class="t m0 x9 h9 y41 ff8 fs4 fc2 sc0 ls0 ws0">v<span class="_ _4"> </span><span class="ff6 fc0">SV<span class="_ _2"></span>M<span class="ff5">的两点过人之处</span></span></div><div class="t m0 xb h9 y42 ff8 fs4 fc3 sc0 ls0 ws0">§<span class="_ _5"> </span><span class="ff5 fc0">推广能力的控制</span></div><div class="t m0 x27 h20 y43 ff6 fs4 fc0 sc0 ls0 ws0">——<span class="ff5">广义最优分类面</span></div><div class="t m0 xb h9 y44 ff8 fs4 fc3 sc0 ls0 ws0">§<span class="_ _5"> </span><span class="ff5 fc0">处理非线性问题的方式</span></div><div class="t m0 x27 h20 y45 ff6 fs4 fc0 sc0 ls0 ws0">——<span class="ff5">核函数</span></div><div class="t m0 x9 h9 y46 ff8 fs4 fc2 sc0 ls0 ws0">v<span class="_ _4"> </span><span class="ff6 fc0">SV<span class="_ _2"></span>M<span class="ff5">核心思想（一）：广义最优分类面</span></span></div><div class="t m0 xb h9 y47 ff8 fs4 fc3 sc0 ls0 ws0">§<span class="_ _5"> </span><span class="ff5 fc0">最优分类面（线）方程</span></div><div class="t m0 xb h9 y48 ff8 fs4 fc3 sc0 ls0 ws0">§<span class="_ _5"> </span><span class="ff5 fc0">最大的分类间隔</span></div><div class="t m0 x27 ha y49 ff5 fs4 fc4 sc0 ls0 ws0">使分类间隔最大实际上就是对推广能力的控制</div><div class="t m0 xb h9 y4a ff8 fs4 fc3 sc0 ls0 ws0">§<span class="_ _5"> </span><span class="ff6 fc0">SV<span class="_ _2"></span>M<span class="ff5">得名由来：支持向量是什么？</span></span></div><div class="t m0 xb h9 y4b ff8 fs4 fc3 sc0 ls0 ws0">§<span class="_ _5"> </span><span class="ff5 fc0">怎么求解广义最有分类面？<span class="ff6">——</span></span></div><div class="t m0 x27 ha y4c ff5 fs4 fc0 sc0 ls0 ws0">这是一个不等式约束下的二次函数寻优问题：</div><div class="t m0 x27 h20 y4d ff5 fs4 fc0 sc0 ls0 ws0">利用<span class="ff6">Lagrange<span class="_ _0"></span><span class="ff5">优<span class="_ _2"></span>化方法将原问题转化为其对偶问题。</span></span></div><div class="c x28 y4e wc h21"><div class="t m0 x29 h22 y4f ff9 fsc fc0 sc0 ls0 ws0">y<span class="_ _8"> </span>w<span class="_ _25"> </span>x<span class="_ _10"> </span>b</div></div><div class="c x2a y50 wd h21"><div class="t m0 x0 h23 y51 ffb fsc fc0 sc0 ls0 ws0">=<span class="_ _a"> </span>⋅<span class="_ _3"> </span>+</div></div><div class="c x2b y52 we h24"><div class="t m0 x0 h25 y53 ff6 fsd fc0 sc0 ls0 ws0">mar<span class="_ _0"></span>gin<span class="_ _26"> </span>2<span class="_ _27"> </span>/</div></div><div class="c x2c y52 wf h24"><div class="t m0 x0 h26 y53 ff9 fsd fc0 sc0 ls0 ws0">w</div></div><div class="t m0 x13 h27 y54 ffb fsd fc0 sc0 ls0 ws0">=</div><div class="c x2d y55 w10 h28"><div class="t m0 x0 h29 y56 ff6 fse fc0 sc0 ls0 ws0">*</div></div><div class="c x2e y57 w11 h2a"><div class="t m0 x0 h2b y58 ff9 fsf fc0 sc0 ls0 ws0">w</div></div></div><div class="pi" data-data='{"ctm":[1.140576,0.000000,0.000000,1.140576,0.000000,0.000000]}'></div></div>