| <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd"> |
| <html xmlns="http://www.w3.org/1999/xhtml"> |
| <head> |
| <meta http-equiv="Content-Type" content="text/xhtml;charset=UTF-8"/> |
| <title>MADlib: online_sv.sql_in Source File</title> |
| |
| <link href="tabs.css" rel="stylesheet" type="text/css"/> |
| <link href="doxygen.css" rel="stylesheet" type="text/css" /> |
| <link href="navtree.css" rel="stylesheet" type="text/css"/> |
| <script type="text/javascript" src="jquery.js"></script> |
| <script type="text/javascript" src="resize.js"></script> |
| <script type="text/javascript" src="navtree.js"></script> |
| <script type="text/javascript"> |
| $(document).ready(initResizable); |
| </script> |
| <link href="search/search.css" rel="stylesheet" type="text/css"/> |
| <script type="text/javascript" src="search/search.js"></script> |
| <script type="text/javascript"> |
| $(document).ready(function() { searchBox.OnSelectItem(0); }); |
| </script> |
| <script src="../mathjax/MathJax.js"> |
| MathJax.Hub.Config({ |
| extensions: ["tex2jax.js", "TeX/AMSmath.js", "TeX/AMSsymbols.js"], |
| jax: ["input/TeX","output/HTML-CSS"], |
| }); |
| </script> |
| </head> |
| <body> |
| <div id="top"><!-- do not remove this div! --> |
| |
| |
| <div id="titlearea"> |
| <table cellspacing="0" cellpadding="0"> |
| <tbody> |
| <tr style="height: 56px;"> |
| |
| |
| <td style="padding-left: 0.5em;"> |
| <div id="projectname">MADlib |
|  <span id="projectnumber">0.6</span> <span style="font-size:10pt; font-style:italic"><a href="../latest/./online__sv_8sql__in_source.html"> A newer version is available</a></span> |
| </div> |
| <div id="projectbrief">User Documentation</div> |
| </td> |
| |
| |
| |
| </tr> |
| </tbody> |
| </table> |
| </div> |
| |
| <!-- Generated by Doxygen 1.7.5.1 --> |
| <script type="text/javascript"> |
| var searchBox = new SearchBox("searchBox", "search",false,'Search'); |
| </script> |
| <script type="text/javascript" src="dynsections.js"></script> |
| <div id="navrow1" class="tabs"> |
| <ul class="tablist"> |
| <li><a href="index.html"><span>Main Page</span></a></li> |
| <li><a href="modules.html"><span>Modules</span></a></li> |
| <li class="current"><a href="files.html"><span>Files</span></a></li> |
| <li> |
| <div id="MSearchBox" class="MSearchBoxInactive"> |
| <span class="left"> |
| <img id="MSearchSelect" src="search/mag_sel.png" |
| onmouseover="return searchBox.OnSearchSelectShow()" |
| onmouseout="return searchBox.OnSearchSelectHide()" |
| alt=""/> |
| <input type="text" id="MSearchField" value="Search" accesskey="S" |
| onfocus="searchBox.OnSearchFieldFocus(true)" |
| onblur="searchBox.OnSearchFieldFocus(false)" |
| onkeyup="searchBox.OnSearchFieldChange(event)"/> |
| </span><span class="right"> |
| <a id="MSearchClose" href="javascript:searchBox.CloseResultsWindow()"><img id="MSearchCloseImg" border="0" src="search/close.png" alt=""/></a> |
| </span> |
| </div> |
| </li> |
| </ul> |
| </div> |
| <div id="navrow2" class="tabs2"> |
| <ul class="tablist"> |
| <li><a href="files.html"><span>File List</span></a></li> |
| <li><a href="globals.html"><span>File Members</span></a></li> |
| </ul> |
| </div> |
| </div> |
| <div id="side-nav" class="ui-resizable side-nav-resizable"> |
| <div id="nav-tree"> |
| <div id="nav-tree-contents"> |
| </div> |
| </div> |
| <div id="splitbar" style="-moz-user-select:none;" |
| class="ui-resizable-handle"> |
| </div> |
| </div> |
| <script type="text/javascript"> |
| initNavTree('online__sv_8sql__in.html',''); |
| </script> |
| <div id="doc-content"> |
| <div class="header"> |
| <div class="headertitle"> |
| <div class="title">online_sv.sql_in</div> </div> |
| </div> |
| <div class="contents"> |
| <a href="online__sv_8sql__in.html">Go to the documentation of this file.</a><div class="fragment"><pre class="fragment"><a name="l00001"></a>00001 <span class="comment">/* ----------------------------------------------------------------------- */</span><span class="comment">/** </span> |
| <a name="l00002"></a>00002 <span class="comment"> *</span> |
| <a name="l00003"></a>00003 <span class="comment"> * @file online_sv.sql_in</span> |
| <a name="l00004"></a>00004 <span class="comment"> *</span> |
| <a name="l00005"></a>00005 <span class="comment"> * @brief SQL functions for support vector machines</span> |
| <a name="l00006"></a>00006 <span class="comment"> * @sa For an introduction to Support vector machines (SVMs) and related kernel</span> |
| <a name="l00007"></a>00007 <span class="comment"> * methods, see the module description \ref grp_kernmach.</span> |
| <a name="l00008"></a>00008 <span class="comment"> *</span> |
| <a name="l00009"></a>00009 <span class="comment"> */</span><span class="comment">/* ------------------------------------------------------------------------*/</span> |
| <a name="l00010"></a>00010 |
| <a name="l00011"></a>00011 m4_include(`SQLCommon.m4<span class="stringliteral">')</span> |
| <a name="l00012"></a>00012 <span class="stringliteral"></span><span class="comment"></span> |
| <a name="l00013"></a>00013 <span class="comment">/**</span> |
| <a name="l00014"></a>00014 <span class="comment">@addtogroup grp_kernmach</span> |
| <a name="l00015"></a>00015 <span class="comment"></span> |
| <a name="l00016"></a>00016 <span class="comment">@about</span> |
| <a name="l00017"></a>00017 <span class="comment"></span> |
| <a name="l00018"></a>00018 <span class="comment">Support vector machines (SVMs) and related kernel methods have been one of </span> |
| <a name="l00019"></a>00019 <span class="comment">the most popular and well-studied machine learning techniques of the </span> |
| <a name="l00020"></a>00020 <span class="comment">past 15 years, with an amazing number of innovations and applications.</span> |
| <a name="l00021"></a>00021 <span class="comment"></span> |
| <a name="l00022"></a>00022 <span class="comment">In a nutshell, an SVM model \f$f(x)\f$ takes the form of</span> |
| <a name="l00023"></a>00023 <span class="comment">\f[</span> |
| <a name="l00024"></a>00024 <span class="comment"> f(x) = \sum_i \alpha_i k(x_i,x),</span> |
| <a name="l00025"></a>00025 <span class="comment">\f]</span> |
| <a name="l00026"></a>00026 <span class="comment">where each \f$ \alpha_i \f$ is a real number, each \f$ \boldsymbol x_i \f$ is a</span> |
| <a name="l00027"></a>00027 <span class="comment">data point from the training set (called a support vector), and</span> |
| <a name="l00028"></a>00028 <span class="comment">\f$ k(\cdot, \cdot) \f$ is a kernel function that measures how "similar" two</span> |
| <a name="l00029"></a>00029 <span class="comment">objects are. In regression, \f$ f(\boldsymbol x) \f$ is the regression function</span> |
| <a name="l00030"></a>00030 <span class="comment">we seek. In classification, \f$ f(\boldsymbol x) \f$ serves as</span> |
| <a name="l00031"></a>00031 <span class="comment">the decision boundary; so for example in binary classification, the predictor </span> |
| <a name="l00032"></a>00032 <span class="comment">can output class 1 for object \f$x\f$ if \f$ f(\boldsymbol x) \geq 0 \f$, and class</span> |
| <a name="l00033"></a>00033 <span class="comment">2 otherwise.</span> |
| <a name="l00034"></a>00034 <span class="comment"></span> |
| <a name="l00035"></a>00035 <span class="comment">In the case when the kernel function \f$ k(\cdot, \cdot) \f$ is the standard</span> |
| <a name="l00036"></a>00036 <span class="comment">inner product on vectors, \f$ f(\boldsymbol x) \f$ is just an alternative way of</span> |
| <a name="l00037"></a>00037 <span class="comment">writing a linear function</span> |
| <a name="l00038"></a>00038 <span class="comment">\f[</span> |
| <a name="l00039"></a>00039 <span class="comment"> f'(\boldsymbol x) = \langle \boldsymbol w, \boldsymbol x \rangle,</span> |
| <a name="l00040"></a>00040 <span class="comment">\f]</span> |
| <a name="l00041"></a>00041 <span class="comment">where \f$ \boldsymbol w \f$ is a weight vector having the same dimension as</span> |
| <a name="l00042"></a>00042 <span class="comment">\f$ \boldsymbol x \f$. One of the key points of SVMs is that we can use more</span> |
| <a name="l00043"></a>00043 <span class="comment">fancy kernel functions to efficiently learn linear models in high-dimensional</span> |
| <a name="l00044"></a>00044 <span class="comment">feature spaces, since \f$ k(\boldsymbol x_i, \boldsymbol x_j) \f$ can be</span> |
| <a name="l00045"></a>00045 <span class="comment">understood as an efficient way of computing an inner product in the feature</span> |
| <a name="l00046"></a>00046 <span class="comment">space:</span> |
| <a name="l00047"></a>00047 <span class="comment">\f[</span> |
| <a name="l00048"></a>00048 <span class="comment"> k(\boldsymbol x_i, \boldsymbol x_j)</span> |
| <a name="l00049"></a>00049 <span class="comment"> = \langle \phi(\boldsymbol x_i), \phi(\boldsymbol x_j) \rangle,</span> |
| <a name="l00050"></a>00050 <span class="comment">\f]</span> |
| <a name="l00051"></a>00051 <span class="comment">where \f$ \phi(\boldsymbol x) \f$ projects \f$ \boldsymbol x \f$ into a</span> |
| <a name="l00052"></a>00052 <span class="comment">(possibly infinite-dimensional) feature space.</span> |
| <a name="l00053"></a>00053 <span class="comment"></span> |
| <a name="l00054"></a>00054 <span class="comment">There are many algorithms for learning kernel machines. This module</span> |
| <a name="l00055"></a>00055 <span class="comment">implements the class of online learning with kernels algorithms</span> |
| <a name="l00056"></a>00056 <span class="comment">described in Kivinen et al. [1]. It also includes the Stochastic</span> |
| <a name="l00057"></a>00057 <span class="comment">Gradient Descent (SGD) method [3] for learning linear SVMs with the Hinge</span> |
| <a name="l00058"></a>00058 <span class="comment">loss \f$l(z) = \max(0, 1-z)\f$. See also the book Scholkopf and Smola [2] for much more</span> |
| <a name="l00059"></a>00059 <span class="comment">details.</span> |
| <a name="l00060"></a>00060 <span class="comment"></span> |
| <a name="l00061"></a>00061 <span class="comment">The SGD implementation is based on L&eacute;on Bottou's SGD package</span> |
| <a name="l00062"></a>00062 <span class="comment">(http://leon.bottou.org/projects/sgd). The methods introduced in [1]</span> |
| <a name="l00063"></a>00063 <span class="comment">are implemented according to their original descriptions, except that</span> |
| <a name="l00064"></a>00064 <span class="comment">we only update the support vector model when we make a significant</span> |
| <a name="l00065"></a>00065 <span class="comment">error. The original algorithms in [1] update the support vector model at</span> |
| <a name="l00066"></a>00066 <span class="comment">every step, even when no error was made, in the name of</span> |
| <a name="l00067"></a>00067 <span class="comment">regularisation. For practical purposes, and this is verified</span> |
| <a name="l00068"></a>00068 <span class="comment">empirically to a certain degree, updating only when necessary is both</span> |
| <a name="l00069"></a>00069 <span class="comment">faster and better from a learning-theoretic point of view, at least in</span> |
| <a name="l00070"></a>00070 <span class="comment">the i.i.d. setting.</span> |
| <a name="l00071"></a>00071 <span class="comment"></span> |
| <a name="l00072"></a>00072 <span class="comment">Methods for classification, regression and novelty detection are </span> |
| <a name="l00073"></a>00073 <span class="comment">available. Multiple instances of the algorithms can be executed </span> |
| <a name="l00074"></a>00074 <span class="comment">in parallel on different subsets of the training data. The resultant</span> |
| <a name="l00075"></a>00075 <span class="comment">support vector models can then be combined using standard techniques</span> |
| <a name="l00076"></a>00076 <span class="comment">like averaging or majority voting.</span> |
| <a name="l00077"></a>00077 <span class="comment"></span> |
| <a name="l00078"></a>00078 <span class="comment">Training data points are accessed via a table or a view. The support</span> |
| <a name="l00079"></a>00079 <span class="comment">vector models can also be stored in tables for fast execution.</span> |
| <a name="l00080"></a>00080 <span class="comment"></span> |
| <a name="l00081"></a>00081 <span class="comment">@input</span> |
| <a name="l00082"></a>00082 <span class="comment">For classification and regression, the training table/view is expected to be of the following form (the array size of <em>ind</em> must not be greater than 102,400.):\n</span> |
| <a name="l00083"></a>00083 <span class="comment"><pre>{TABLE|VIEW} <em>input_table</em> (</span> |
| <a name="l00084"></a>00084 <span class="comment"> ...</span> |
| <a name="l00085"></a>00085 <span class="comment"> <em>id</em> INT,</span> |
| <a name="l00086"></a>00086 <span class="comment"> <em>ind</em> FLOAT8[],</span> |
| <a name="l00087"></a>00087 <span class="comment"> <em>label</em> FLOAT8,</span> |
| <a name="l00088"></a>00088 <span class="comment"> ...</span> |
| <a name="l00089"></a>00089 <span class="comment">)</pre></span> |
| <a name="l00090"></a>00090 <span class="comment">For novelty detection, the label field is not required.</span> |
| <a name="l00091"></a>00091 <span class="comment"></span> |
| <a name="l00092"></a>00092 <span class="comment">@usage</span> |
| <a name="l00093"></a>00093 <span class="comment"></span> |
| <a name="l00094"></a>00094 <span class="comment">- Regression learning is achieved through the following function:</span> |
| <a name="l00095"></a>00095 <span class="comment"> <pre>SELECT \ref svm_regression(</span> |
| <a name="l00096"></a>00096 <span class="comment"> '<em>input_table</em>', '<em>model_table</em>', <em>parallel</em>, '<em>kernel_func</em>', </span> |
| <a name="l00097"></a>00097 <span class="comment"> <em>verbose DEFAULT false</em>, <em>eta DEFAULT 0.1</em>, <em>nu DEFAULT 0.005</em>, <em>slambda DEFAULT 0.05</em></span> |
| <a name="l00098"></a>00098 <span class="comment"> );</pre> </span> |
| <a name="l00099"></a>00099 <span class="comment"></span> |
| <a name="l00100"></a>00100 <span class="comment">- Classification learning is achieved through the following two</span> |
| <a name="l00101"></a>00101 <span class="comment"> functions:</span> |
| <a name="l00102"></a>00102 <span class="comment"> -# Learn linear SVM(s) using SGD [3]:</span> |
| <a name="l00103"></a>00103 <span class="comment"> <pre>SELECT \ref lsvm_classification(</span> |
| <a name="l00104"></a>00104 <span class="comment"> '<em>input_table</em>', '<em>model_table</em>', <em>parallel</em>, </span> |
| <a name="l00105"></a>00105 <span class="comment"> <em>verbose DEFAULT false</em>, <em>eta DEFAULT 0.1</em>, <em>reg DEFAULT 0.001</em></span> |
| <a name="l00106"></a>00106 <span class="comment"> );</pre> </span> |
| <a name="l00107"></a>00107 <span class="comment"> -# Learn linear or non-linear SVM(s) using the method described in [1]:</span> |
| <a name="l00108"></a>00108 <span class="comment"> <pre>SELECT \ref svm_classification(</span> |
| <a name="l00109"></a>00109 <span class="comment"> '<em>input_table</em>', '<em>model_table</em>', <em>parallel</em>, '<em>kernel_func</em>', </span> |
| <a name="l00110"></a>00110 <span class="comment"> <em>verbose DEFAULT false</em>, <em>eta DEFAULT 0.1</em>, <em>nu DEFAULT 0.005</em></span> |
| <a name="l00111"></a>00111 <span class="comment"> );</pre> </span> |
| <a name="l00112"></a>00112 <span class="comment"></span> |
| <a name="l00113"></a>00113 <span class="comment">- Novelty detection is achieved through the following function:</span> |
| <a name="l00114"></a>00114 <span class="comment"> <pre>SELECT \ref svm_novelty_detection(</span> |
| <a name="l00115"></a>00115 <span class="comment"> '<em>input_table</em>', '<em>model_table</em>', <em>parallel</em>, '<em>kernel_func</em>', </span> |
| <a name="l00116"></a>00116 <span class="comment"> <em>verbose DEFAULT false</em>, <em>eta DEFAULT 0.1</em>, <em>nu DEFAULT 0.005</em></span> |
| <a name="l00117"></a>00117 <span class="comment"> );</pre></span> |
| <a name="l00118"></a>00118 <span class="comment"> Assuming the model_table parameter takes on value 'model', each learning function will produce two tables </span> |
| <a name="l00119"></a>00119 <span class="comment"> as output: 'model' and 'model_param'.</span> |
| <a name="l00120"></a>00120 <span class="comment"> The first contains the support vectors of the model(s) learned.</span> |
| <a name="l00121"></a>00121 <span class="comment"> The second contains the parameters of the model(s) learned, which includes information like the kernel function</span> |
| <a name="l00122"></a>00122 <span class="comment"> used and the value of the intercept, if there is one.</span> |
| <a name="l00123"></a>00123 <span class="comment"></span> |
| <a name="l00124"></a>00124 <span class="comment">- To make predictions on a single data point x using a single model</span> |
| <a name="l00125"></a>00125 <span class="comment"> learned previously, we use the function</span> |
| <a name="l00126"></a>00126 <span class="comment"> <pre>SELECT \ref</span> |
| <a name="l00127"></a>00127 <span class="comment"> svm_predict('<em>model_table</em>',<em>x</em>);</pre></span> |
| <a name="l00128"></a>00128 <span class="comment"> If the model is produced by the lsvm_classification() function, use</span> |
| <a name="l00129"></a>00129 <span class="comment"> the following prediction function instead</span> |
| <a name="l00130"></a>00130 <span class="comment"> <pre>SELECT \ref</span> |
| <a name="l00131"></a>00131 <span class="comment"> lsvm_predict('<em>model_table</em>',<em>x</em>);</pre></span> |
| <a name="l00132"></a>00132 <span class="comment"></span> |
| <a name="l00133"></a>00133 <span class="comment">- To make predictions on new data points using multiple models</span> |
| <a name="l00134"></a>00134 <span class="comment"> learned in parallel, we use the function</span> |
| <a name="l00135"></a>00135 <span class="comment"> <pre>SELECT \ref</span> |
| <a name="l00136"></a>00136 <span class="comment"> svm_predict_combo('<em>model_table</em>',<em>x</em>);</pre></span> |
| <a name="l00137"></a>00137 <span class="comment"> If the models are produced by the lsvm_classification() function, use</span> |
| <a name="l00138"></a>00138 <span class="comment"> the following prediction function instead</span> |
| <a name="l00139"></a>00139 <span class="comment"> <pre>SELECT \ref</span> |
| <a name="l00140"></a>00140 <span class="comment"> lsvm_predict_combo('<em>model_table</em>',<em>x</em>);</pre></span> |
| <a name="l00141"></a>00141 <span class="comment"></span> |
| <a name="l00142"></a>00142 <span class="comment"></span> |
| <a name="l00143"></a>00143 <span class="comment">- Note that, at the moment, we cannot use MADLIB_SCHEMA.svm_predict() and MADLIB_SCHEMA.svm_predict_combo()</span> |
| <a name="l00144"></a>00144 <span class="comment"> on multiple data points. For example, something like the following will fail:</span> |
| <a name="l00145"></a>00145 <span class="comment"> <pre>SELECT \ref svm_predict('<em>model_table</em>',<em>x</em>) FROM data_table;</pre></span> |
| <a name="l00146"></a>00146 <span class="comment"> Instead, to make predictions on new data points stored in a table using</span> |
| <a name="l00147"></a>00147 <span class="comment"> previously learned models, we use the function:</span> |
| <a name="l00148"></a>00148 <span class="comment"> <pre>SELECT \ref svm_predict_batch('<em>input_table</em>', '<em>data_col</em>', '<em>id_col</em>', '<em>model_table</em>', '<em>output_table</em>', <em>parallel</em>);</pre></span> |
| <a name="l00149"></a>00149 <span class="comment"> The output_table is created during the function call; an existing table with </span> |
| <a name="l00150"></a>00150 <span class="comment"> the same name will be dropped.</span> |
| <a name="l00151"></a>00151 <span class="comment"> If the parallel parameter is true, then each data point in the input table will have multiple </span> |
| <a name="l00152"></a>00152 <span class="comment"> predicted values corresponding to the number of models learned in</span> |
| <a name="l00153"></a>00153 <span class="comment"> parallel.\n\n</span> |
| <a name="l00154"></a>00154 <span class="comment"> Similarly, use the following function for batch prediction if the</span> |
| <a name="l00155"></a>00155 <span class="comment"> model(s) is produced by the lsvm_classification() function:</span> |
| <a name="l00156"></a>00156 <span class="comment"> <pre>SELECT \ref lsvm_predict_batch('<em>input_table</em>', '<em>data_col</em>', '<em>id_col</em>', '<em>model_table</em>','<em>output_table</em>', <em>parallel</em>);</pre></span> |
| <a name="l00157"></a>00157 <span class="comment"> </span> |
| <a name="l00158"></a>00158 <span class="comment"> </span> |
| <a name="l00159"></a>00159 <span class="comment"></span> |
| <a name="l00160"></a>00160 <span class="comment">@implementation</span> |
| <a name="l00161"></a>00161 <span class="comment"></span> |
| <a name="l00162"></a>00162 <span class="comment">Currently, three kernel functions have been implemented: dot product (\ref svm_dot), polynomial (\ref svm_polynomial) and Gaussian (\ref svm_gaussian) kernels. To use the dot product kernel function,</span> |
| <a name="l00163"></a>00163 <span class="comment">simply use '<tt><em>MADLIB_SCHEMA.svm_dot</em></tt>' as the <tt>kernel_func</tt> argument, which accepts any function that takes in two float[] and returns a float. To use the polynomial or Gaussian kernels,</span> |
| <a name="l00164"></a>00164 <span class="comment">a wrapper function is needed since these kernels require additional input parameters (see online_sv.sql_in for input parameters).</span> |
| <a name="l00165"></a>00165 <span class="comment"></span> |
| <a name="l00166"></a>00166 <span class="comment">For example, to use the polynomial kernel with degree 2, first create a wrapper function:</span> |
| <a name="l00167"></a>00167 <span class="comment"><pre>CREATE OR REPLACE FUNCTION mykernel(FLOAT[],FLOAT[]) RETURNS FLOAT AS $$</span> |
| <a name="l00168"></a>00168 <span class="comment"> SELECT \ref svm_polynomial($1,$2,2)</span> |
| <a name="l00169"></a>00169 <span class="comment">$$ language sql;</pre></span> |
| <a name="l00170"></a>00170 <span class="comment">Then call the SVM learning functions with <tt>mykernel</tt> as the argument to <tt>kernel_func</tt>.</span> |
| <a name="l00171"></a>00171 <span class="comment"><pre>SELECT \ref svm_regression('my_schema.my_train_data', 'mymodel', false, 'mykernel');</pre></span> |
| <a name="l00172"></a>00172 <span class="comment"></span> |
| <a name="l00173"></a>00173 <span class="comment">To drop all tables pertaining to the model, we can use</span> |
| <a name="l00174"></a>00174 <span class="comment"><pre>SELECT \ref svm_drop_model('model_table');</pre></span> |
| <a name="l00175"></a>00175 <span class="comment"></span> |
| <a name="l00176"></a>00176 <span class="comment">@examp</span> |
| <a name="l00177"></a>00177 <span class="comment"></span> |
| <a name="l00178"></a>00178 <span class="comment">As a general first step, we need to prepare and populate an input </span> |
| <a name="l00179"></a>00179 <span class="comment">table/view with the following structure:</span> |
| <a name="l00180"></a>00180 <span class="comment">\code </span> |
| <a name="l00181"></a>00181 <span class="comment">TABLE/VIEW my_schema.my_input_table </span> |
| <a name="l00182"></a>00182 <span class="comment">( </span> |
| <a name="l00183"></a>00183 <span class="comment"> id INT, -- point ID</span> |
| <a name="l00184"></a>00184 <span class="comment"> ind FLOAT8[], -- data point</span> |
| <a name="l00185"></a>00185 <span class="comment"> label FLOAT8 -- label of data point</span> |
| <a name="l00186"></a>00186 <span class="comment">);</span> |
| <a name="l00187"></a>00187 <span class="comment">\endcode </span> |
| <a name="l00188"></a>00188 <span class="comment"> Note: The label field is not required for novelty detection.</span> |
| <a name="l00189"></a>00189 <span class="comment"> </span> |
| <a name="l00190"></a>00190 <span class="comment"></span> |
| <a name="l00191"></a>00191 <span class="comment"><strong>Example usage for regression</strong>:</span> |
| <a name="l00192"></a>00192 <span class="comment"> -# We can randomly generate 1000 5-dimensional data labelled by the simple target function </span> |
| <a name="l00193"></a>00193 <span class="comment">\code</span> |
| <a name="l00194"></a>00194 <span class="comment">t(x) = if x[5] = 10 then 50 else if x[5] = -10 then 50 else 0;</span> |
| <a name="l00195"></a>00195 <span class="comment">\endcode</span> |
| <a name="l00196"></a>00196 <span class="comment">and store that in the my_schema.my_train_data table as follows:</span> |
| <a name="l00197"></a>00197 <span class="comment">\code</span> |
| <a name="l00198"></a>00198 <span class="comment">sql> select MADLIB_SCHEMA.svm_generate_reg_data('my_schema.my_train_data', 1000, 5);</span> |
| <a name="l00199"></a>00199 <span class="comment">\endcode</span> |
| <a name="l00200"></a>00200 <span class="comment"> -# We can now learn a regression model and store the resultant model</span> |
| <a name="l00201"></a>00201 <span class="comment"> under the name 'myexp'.</span> |
| <a name="l00202"></a>00202 <span class="comment">\code</span> |
| <a name="l00203"></a>00203 <span class="comment">sql> select MADLIB_SCHEMA.svm_regression('my_schema.my_train_data', 'myexp', false, 'MADLIB_SCHEMA.svm_dot');</span> |
| <a name="l00204"></a>00204 <span class="comment">\endcode</span> |
| <a name="l00205"></a>00205 <span class="comment"> -# We can now start using it to predict the labels of new data points </span> |
| <a name="l00206"></a>00206 <span class="comment"> like as follows:</span> |
| <a name="l00207"></a>00207 <span class="comment">\code</span> |
| <a name="l00208"></a>00208 <span class="comment">sql> select MADLIB_SCHEMA.svm_predict('myexp', '{1,2,4,20,10}');</span> |
| <a name="l00209"></a>00209 <span class="comment">sql> select MADLIB_SCHEMA.svm_predict('myexp', '{1,2,4,20,-10}');</span> |
| <a name="l00210"></a>00210 <span class="comment">\endcode</span> |
| <a name="l00211"></a>00211 <span class="comment"> -# To learn multiple support vector models, we replace the learning step above by </span> |
| <a name="l00212"></a>00212 <span class="comment">\code</span> |
| <a name="l00213"></a>00213 <span class="comment">sql> select MADLIB_SCHEMA.svm_regression('my_schema.my_train_data', 'myexp', true, 'MADLIB_SCHEMA.svm_dot');</span> |
| <a name="l00214"></a>00214 <span class="comment">\endcode</span> |
| <a name="l00215"></a>00215 <span class="comment">The resultant models can be used for prediction as follows:</span> |
| <a name="l00216"></a>00216 <span class="comment">\code</span> |
| <a name="l00217"></a>00217 <span class="comment">sql> select * from MADLIB_SCHEMA.svm_predict_combo('myexp', '{1,2,4,20,10}');</span> |
| <a name="l00218"></a>00218 <span class="comment">\endcode</span> |
| <a name="l00219"></a>00219 <span class="comment"> -# We can also predict the labels of all the data points stored in a table.</span> |
| <a name="l00220"></a>00220 <span class="comment"> For example, we can execute the following:</span> |
| <a name="l00221"></a>00221 <span class="comment">\code</span> |
| <a name="l00222"></a>00222 <span class="comment">sql> create table MADLIB_SCHEMA.svm_reg_test ( id int, ind float8[] );</span> |
| <a name="l00223"></a>00223 <span class="comment">sql> insert into MADLIB_SCHEMA.svm_reg_test (select id, ind from my_schema.my_train_data limit 20);</span> |
| <a name="l00224"></a>00224 <span class="comment">sql> select MADLIB_SCHEMA.svm_predict_batch('MADLIB_SCHEMA.svm_reg_test', 'ind', 'id', 'myexp', 'MADLIB_SCHEMA.svm_reg_output1', false); </span> |
| <a name="l00225"></a>00225 <span class="comment">sql> select * from MADLIB_SCHEMA.svm_reg_output1;</span> |
| <a name="l00226"></a>00226 <span class="comment">sql> select MADLIB_SCHEMA.svm_predict_batch('MADLIB_SCHEMA.svm_reg_test', 'ind', 'id, 'myexp', 'MADLIB_SCHEMA.svm_reg_output2', true);</span> |
| <a name="l00227"></a>00227 <span class="comment">sql> select * from MADLIB_SCHEMA.svm_reg_output2;</span> |
| <a name="l00228"></a>00228 <span class="comment">\endcode </span> |
| <a name="l00229"></a>00229 <span class="comment"></span> |
| <a name="l00230"></a>00230 <span class="comment"><strong>Example usage for classification:</strong></span> |
| <a name="l00231"></a>00231 <span class="comment">-# We can randomly generate 2000 5-dimensional data labelled by the simple</span> |
| <a name="l00232"></a>00232 <span class="comment">target function </span> |
| <a name="l00233"></a>00233 <span class="comment">\code</span> |
| <a name="l00234"></a>00234 <span class="comment">t(x) = if x[1] > 0 and x[2] < 0 then 1 else -1;</span> |
| <a name="l00235"></a>00235 <span class="comment">\endcode</span> |
| <a name="l00236"></a>00236 <span class="comment">and store that in the my_schema.my_train_data table as follows:</span> |
| <a name="l00237"></a>00237 <span class="comment">\code </span> |
| <a name="l00238"></a>00238 <span class="comment">sql> select MADLIB_SCHEMA.svm_generate_cls_data('my_schema.my_train_data', 2000, 5);</span> |
| <a name="l00239"></a>00239 <span class="comment">\endcode</span> |
| <a name="l00240"></a>00240 <span class="comment">-# We can now learn a classification model and store the resultant model</span> |
| <a name="l00241"></a>00241 <span class="comment">under the name 'myexpc'.</span> |
| <a name="l00242"></a>00242 <span class="comment">\code</span> |
| <a name="l00243"></a>00243 <span class="comment">sql> select MADLIB_SCHEMA.svm_classification('my_schema.my_train_data', 'myexpc', false, 'MADLIB_SCHEMA.svm_dot');</span> |
| <a name="l00244"></a>00244 <span class="comment">\endcode</span> |
| <a name="l00245"></a>00245 <span class="comment">-# We can now start using it to predict the labels of new data points </span> |
| <a name="l00246"></a>00246 <span class="comment">like as follows:</span> |
| <a name="l00247"></a>00247 <span class="comment">\code</span> |
| <a name="l00248"></a>00248 <span class="comment">sql> select MADLIB_SCHEMA.svm_predict('myexpc', '{10,-2,4,20,10}');</span> |
| <a name="l00249"></a>00249 <span class="comment">\endcode </span> |
| <a name="l00250"></a>00250 <span class="comment">-# To learn multiple support vector models, replace the model-building and prediction steps above by </span> |
| <a name="l00251"></a>00251 <span class="comment">\code</span> |
| <a name="l00252"></a>00252 <span class="comment">sql> select MADLIB_SCHEMA.svm_classification('my_schema.my_train_data', 'myexpc', true, 'MADLIB_SCHEMA.svm_dot');</span> |
| <a name="l00253"></a>00253 <span class="comment">sql> select * from MADLIB_SCHEMA.svm_predict_combo('myexpc', '{10,-2,4,20,10}');</span> |
| <a name="l00254"></a>00254 <span class="comment">\endcode</span> |
| <a name="l00255"></a>00255 <span class="comment">-# To learn a linear support vector model using SGD, replace the model-building and prediction steps above by </span> |
| <a name="l00256"></a>00256 <span class="comment">\code</span> |
| <a name="l00257"></a>00257 <span class="comment">sql> select MADLIB_SCHEMA.lsvm_classification('my_schema.my_train_data', 'myexpc', false);</span> |
| <a name="l00258"></a>00258 <span class="comment">sql> select MADLIB_SCHEMA.lsvm_predict('myexpc', '{10,-2,4,20,10}');</span> |
| <a name="l00259"></a>00259 <span class="comment">\endcode</span> |
| <a name="l00260"></a>00260 <span class="comment">-# To learn multiple linear support vector models using SGD, replace the model-building and prediction steps above by </span> |
| <a name="l00261"></a>00261 <span class="comment">\code</span> |
| <a name="l00262"></a>00262 <span class="comment">sql> select MADLIB_SCHEMA.lsvm_classification('my_schema.my_train_data', 'myexpc', true);</span> |
| <a name="l00263"></a>00263 <span class="comment">sql> select MADLIB_SCHEMA.lsvm_predict_combo('myexpc', '{10,-2,4,20,10}');</span> |
| <a name="l00264"></a>00264 <span class="comment">\endcode</span> |
| <a name="l00265"></a>00265 <span class="comment"></span> |
| <a name="l00266"></a>00266 <span class="comment"><strong>Example usage for novelty detection:</strong></span> |
| <a name="l00267"></a>00267 <span class="comment">-# We can randomly generate 100 2-dimensional data (the normal cases)</span> |
| <a name="l00268"></a>00268 <span class="comment">and store that in the my_schema.my_train_data table as follows:</span> |
| <a name="l00269"></a>00269 <span class="comment">\code</span> |
| <a name="l00270"></a>00270 <span class="comment">sql> select MADLIB_SCHEMA.svm_generate_nd_data('my_schema.my_train_data', 100, 2);</span> |
| <a name="l00271"></a>00271 <span class="comment">\endcode</span> |
| <a name="l00272"></a>00272 <span class="comment">-# Learning and predicting using a single novelty detection model can be done as follows:</span> |
| <a name="l00273"></a>00273 <span class="comment">\code</span> |
| <a name="l00274"></a>00274 <span class="comment">sql> select MADLIB_SCHEMA.svm_novelty_detection('my_schema.my_train_data', 'myexpnd', false, 'MADLIB_SCHEMA.svm_dot');</span> |
| <a name="l00275"></a>00275 <span class="comment">sql> select MADLIB_SCHEMA.svm_predict('myexpnd', '{10,-10}'); </span> |
| <a name="l00276"></a>00276 <span class="comment">sql> select MADLIB_SCHEMA.svm_predict('myexpnd', '{-1,-1}'); </span> |
| <a name="l00277"></a>00277 <span class="comment">\endcode</span> |
| <a name="l00278"></a>00278 <span class="comment">-# Learning and predicting using multiple models can be done as follows:</span> |
| <a name="l00279"></a>00279 <span class="comment">\code</span> |
| <a name="l00280"></a>00280 <span class="comment">sql> select MADLIB_SCHEMA.svm_novelty_detection('my_schema.my_train_data', 'myexpnd', true, 'MADLIB_SCHEMA.svm_dot');</span> |
| <a name="l00281"></a>00281 <span class="comment">sql> select * from MADLIB_SCHEMA.svm_predict_combo('myexpnd', '{10,-10}'); </span> |
| <a name="l00282"></a>00282 <span class="comment">sql> select * from MADLIB_SCHEMA.svm_predict_combo('myexpnd', '{-1,-1}'); </span> |
| <a name="l00283"></a>00283 <span class="comment">\endcode</span> |
| <a name="l00284"></a>00284 <span class="comment"></span> |
| <a name="l00285"></a>00285 <span class="comment"></span> |
| <a name="l00286"></a>00286 <span class="comment">@literature</span> |
| <a name="l00287"></a>00287 <span class="comment"></span> |
| <a name="l00288"></a>00288 <span class="comment">[1] Jyrki Kivinen, Alexander J. Smola, and Robert C. Williamson: <em>Online</span> |
| <a name="l00289"></a>00289 <span class="comment"> Learning with Kernels</em>, IEEE Transactions on Signal Processing, 52(8),</span> |
| <a name="l00290"></a>00290 <span class="comment"> 2165-2176, 2004.</span> |
| <a name="l00291"></a>00291 <span class="comment"></span> |
| <a name="l00292"></a>00292 <span class="comment">[2] Bernhard Scholkopf and Alexander J. Smola: <em>Learning with Kernels:</span> |
| <a name="l00293"></a>00293 <span class="comment"> Support Vector Machines, Regularization, Optimization, and Beyond</em>, </span> |
| <a name="l00294"></a>00294 <span class="comment"> MIT Press, 2002.</span> |
| <a name="l00295"></a>00295 <span class="comment"></span> |
| <a name="l00296"></a>00296 <span class="comment">[3] L&eacute;on Bottou: <em>Large-Scale Machine Learning with Stochastic</span> |
| <a name="l00297"></a>00297 <span class="comment">Gradient Descent</em>, Proceedings of the 19th International</span> |
| <a name="l00298"></a>00298 <span class="comment">Conference on Computational Statistics, Springer, 2010.</span> |
| <a name="l00299"></a>00299 <span class="comment"> </span> |
| <a name="l00300"></a>00300 <span class="comment">@sa File online_sv.sql_in documenting the SQL functions.</span> |
| <a name="l00301"></a>00301 <span class="comment"></span> |
| <a name="l00302"></a>00302 <span class="comment">@internal</span> |
| <a name="l00303"></a>00303 <span class="comment">@sa namespace online_sv (documenting the implementation in Python)</span> |
| <a name="l00304"></a>00304 <span class="comment">@endinternal</span> |
| <a name="l00305"></a>00305 <span class="comment"> </span> |
| <a name="l00306"></a>00306 <span class="comment">*/</span> |
| <a name="l00307"></a>00307 |
| <a name="l00308"></a>00308 |
| <a name="l00309"></a>00309 |
| <a name="l00310"></a>00310 -- The following is the structure to record the results of a learning process. |
| <a name="l00311"></a>00311 -- We work with arrays of float8 for now; we'll extend the code to work with sparse vectors next. |
| <a name="l00312"></a>00312 -- |
| <a name="l00313"></a>00313 CREATE TYPE MADLIB_SCHEMA.svm_model_rec AS ( |
| <a name="l00314"></a>00314 inds <span class="keywordtype">int</span>, -- number of individuals processed |
| <a name="l00315"></a>00315 cum_err float8, -- cumulative error |
| <a name="l00316"></a>00316 epsilon float8, -- the size of the epsilon tube around the hyperplane, adaptively adjusted by algorithm |
| <a name="l00317"></a>00317 rho float8, -- classification margin |
| <a name="l00318"></a>00318 b float8, -- classifier offset |
| <a name="l00319"></a>00319 nsvs <span class="keywordtype">int</span>, -- number of support vectors |
| <a name="l00320"></a>00320 ind_dim <span class="keywordtype">int</span>, -- the dimension of the individuals |
| <a name="l00321"></a>00321 weights float8[], -- the weight of the support vectors |
| <a name="l00322"></a>00322 individuals float8[], -- the array of support vectors, represented as a 1-D array |
| <a name="l00323"></a>00323 kernel_oid oid -- OID of kernel <span class="keyword">function</span> |
| <a name="l00324"></a>00324 ); |
| <a name="l00325"></a>00325 |
| <a name="l00326"></a>00326 -- The following is the structure to record the results of the linear SVM sgd algorithm |
| <a name="l00327"></a>00327 -- |
| <a name="l00328"></a>00328 CREATE TYPE MADLIB_SCHEMA.lsvm_sgd_model_rec AS ( |
| <a name="l00329"></a>00329 weights float8[], -- the weight vector |
| <a name="l00330"></a>00330 wdiv float8, -- scaling factor <span class="keywordflow">for</span> the weights |
| <a name="l00331"></a>00331 wbias float8, -- offset/bias of the linear model |
| <a name="l00332"></a>00332 ind_dim <span class="keywordtype">int</span>, -- the dimension of the individuals |
| <a name="l00333"></a>00333 inds <span class="keywordtype">int</span>, -- number of individuals processed |
| <a name="l00334"></a>00334 cum_err <span class="keywordtype">int</span> -- cumulative error |
| <a name="l00335"></a>00335 ); |
| <a name="l00336"></a>00336 |
| <a name="l00337"></a>00337 |
| <a name="l00338"></a>00338 -- The following is the <span class="keywordflow">return</span> type of a regression learning process |
| <a name="l00339"></a>00339 -- |
| <a name="l00340"></a>00340 CREATE TYPE MADLIB_SCHEMA.svm_reg_result AS ( |
| <a name="l00341"></a>00341 model_table text, -- table where the model is stored |
| <a name="l00342"></a>00342 model_name text, -- model name |
| <a name="l00343"></a>00343 inds <span class="keywordtype">int</span>, -- number of individuals processed |
| <a name="l00344"></a>00344 cum_err float8, -- cumulative error |
| <a name="l00345"></a>00345 epsilon float8, -- the size of the epsilon tube around the hyperplane, adaptively adjusted by algorithm |
| <a name="l00346"></a>00346 b float8, -- classifier offset |
| <a name="l00347"></a>00347 nsvs <span class="keywordtype">int</span> -- number of support vectors |
| <a name="l00348"></a>00348 ); |
| <a name="l00349"></a>00349 |
| <a name="l00350"></a>00350 -- The following is the <span class="keywordflow">return</span> type of a classification learning process |
| <a name="l00351"></a>00351 -- |
| <a name="l00352"></a>00352 CREATE TYPE MADLIB_SCHEMA.svm_cls_result AS ( |
| <a name="l00353"></a>00353 model_table text, -- table where the model is stored |
| <a name="l00354"></a>00354 model_name text, -- model name |
| <a name="l00355"></a>00355 inds <span class="keywordtype">int</span>, -- number of individuals processed |
| <a name="l00356"></a>00356 cum_err float8, -- cumulative error |
| <a name="l00357"></a>00357 rho float8, -- classification margin |
| <a name="l00358"></a>00358 b float8, -- classifier offset |
| <a name="l00359"></a>00359 nsvs <span class="keywordtype">int</span> -- number of support vectors |
| <a name="l00360"></a>00360 ); |
| <a name="l00361"></a>00361 |
| <a name="l00362"></a>00362 -- The following is the <span class="keywordflow">return</span> type of a linear classifier learning process |
| <a name="l00363"></a>00363 -- |
| <a name="l00364"></a>00364 CREATE TYPE MADLIB_SCHEMA.lsvm_sgd_result AS ( |
| <a name="l00365"></a>00365 model_table text, -- table where the model is stored |
| <a name="l00366"></a>00366 model_name text, -- model name |
| <a name="l00367"></a>00367 inds <span class="keywordtype">int</span>, -- number of individuals processed |
| <a name="l00368"></a>00368 ind_dim <span class="keywordtype">int</span>, -- the dimension of the individuals |
| <a name="l00369"></a>00369 cum_err float8, -- cumulative error |
| <a name="l00370"></a>00370 wdiv float8, -- scaling factor <span class="keywordflow">for</span> the weights |
| <a name="l00371"></a>00371 wbias float8 -- classifier offset |
| <a name="l00372"></a>00372 ); |
| <a name="l00373"></a>00373 |
| <a name="l00374"></a>00374 -- The following is the <span class="keywordflow">return</span> type of a novelty detection learning process |
| <a name="l00375"></a>00375 -- |
| <a name="l00376"></a>00376 CREATE TYPE MADLIB_SCHEMA.svm_nd_result AS ( |
| <a name="l00377"></a>00377 model_table text, -- table where the model is stored |
| <a name="l00378"></a>00378 model_name text, -- model name |
| <a name="l00379"></a>00379 inds <span class="keywordtype">int</span>, -- number of individuals processed |
| <a name="l00380"></a>00380 rho float8, -- classification margin |
| <a name="l00381"></a>00381 nsvs <span class="keywordtype">int</span> -- number of support vectors |
| <a name="l00382"></a>00382 ); |
| <a name="l00383"></a>00383 |
| <a name="l00384"></a>00384 -- The type <span class="keywordflow">for</span> representing support vectors |
| <a name="l00385"></a>00385 -- |
| <a name="l00386"></a>00386 CREATE TYPE MADLIB_SCHEMA.svm_support_vector AS ( <span class="keywordtype">id</span> text, weight float8, sv float8[] ); |
| <a name="l00387"></a>00387 |
| <a name="l00388"></a>00388 |
| <a name="l00389"></a>00389 |
| <a name="l00390"></a>00390 -- Kernel functions are a generalisation of inner products. |
| <a name="l00391"></a>00391 -- They provide the means by which we can extend linear machines to work in non-linear transformed feature spaces. |
| <a name="l00392"></a>00392 -- Here are a few standard kernels: dot product, polynomial kernel, Gaussian kernel. |
| <a name="l00393"></a>00393 --<span class="comment"></span> |
| <a name="l00394"></a>00394 <span class="comment">/**</span> |
| <a name="l00395"></a>00395 <span class="comment"> * @brief Dot product kernel function</span> |
| <a name="l00396"></a>00396 <span class="comment"> *</span> |
| <a name="l00397"></a>00397 <span class="comment"> * @param x The data point \f$ \boldsymbol x \f$</span> |
| <a name="l00398"></a>00398 <span class="comment"> * @param y The data point \f$ \boldsymbol y \f$</span> |
| <a name="l00399"></a>00399 <span class="comment"> * @return Returns dot product of the two data points.</span> |
| <a name="l00400"></a>00400 <span class="comment"> * </span> |
| <a name="l00401"></a>00401 <span class="comment"> */</span> |
| <a name="l00402"></a>00402 CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svm_dot(x float8[], y float8[]) RETURNS float8 |
| <a name="l00403"></a>00403 AS <span class="stringliteral">'MODULE_PATHNAME'</span>, <span class="stringliteral">'svm_dot'</span> LANGUAGE C IMMUTABLE STRICT; |
| <a name="l00404"></a>00404 <span class="comment"></span> |
| <a name="l00405"></a>00405 <span class="comment">/**</span> |
| <a name="l00406"></a>00406 <span class="comment"> * @brief Polynomial kernel function</span> |
| <a name="l00407"></a>00407 <span class="comment"> *</span> |
| <a name="l00408"></a>00408 <span class="comment"> * @param x The data point \f$ \boldsymbol x \f$</span> |
| <a name="l00409"></a>00409 <span class="comment"> * @param y The data point \f$ \boldsymbol y \f$</span> |
| <a name="l00410"></a>00410 <span class="comment"> * @param degree The degree \f$ d \f$</span> |
| <a name="l00411"></a>00411 <span class="comment"> * @return Returns \f$ K(\boldsymbol x,\boldsymbol y)=(\boldsymbol x \cdot \boldsymbol y)^d \f$</span> |
| <a name="l00412"></a>00412 <span class="comment"> * </span> |
| <a name="l00413"></a>00413 <span class="comment"> */</span> |
| <a name="l00414"></a>00414 CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svm_polynomial(x float8[], y float8[], degree float8) RETURNS float8 |
| <a name="l00415"></a>00415 AS <span class="stringliteral">'MODULE_PATHNAME'</span>, <span class="stringliteral">'svm_polynomial'</span> LANGUAGE C IMMUTABLE STRICT; |
| <a name="l00416"></a>00416 <span class="comment"></span> |
| <a name="l00417"></a>00417 <span class="comment">/**</span> |
| <a name="l00418"></a>00418 <span class="comment"> * @brief Gaussian kernel function</span> |
| <a name="l00419"></a>00419 <span class="comment"> *</span> |
| <a name="l00420"></a>00420 <span class="comment"> * @param x The data point \f$ \boldsymbol x \f$</span> |
| <a name="l00421"></a>00421 <span class="comment"> * @param y The data point \f$ \boldsymbol y \f$</span> |
| <a name="l00422"></a>00422 <span class="comment"> * @param gamma The spread \f$ \gamma \f$</span> |
| <a name="l00423"></a>00423 <span class="comment"> * @return Returns \f$ K(\boldsymbol x,\boldsymbol y)=exp(-\gamma || \boldsymbol x \cdot \boldsymbol y ||^2 ) \f$</span> |
| <a name="l00424"></a><a class="code" href="online__sv_8sql__in.html#acc2d778a8eb48ab775ff9c1dff4a3141">00424</a> <span class="comment"> * </span> |
| <a name="l00425"></a>00425 <span class="comment"> */</span> |
| <a name="l00426"></a>00426 CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svm_gaussian(x float8[], y float8[], gamma float8) RETURNS float8 |
| <a name="l00427"></a>00427 AS <span class="stringliteral">'MODULE_PATHNAME'</span>, <span class="stringliteral">'svm_gaussian'</span> LANGUAGE C IMMUTABLE STRICT; |
| <a name="l00428"></a>00428 |
| <a name="l00429"></a>00429 CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svm_predict_sub(<span class="keywordtype">int</span>,<span class="keywordtype">int</span>,float8[],float8[],float8[],text) RETURNS float8 |
| <a name="l00430"></a>00430 AS <span class="stringliteral">'MODULE_PATHNAME'</span>, <span class="stringliteral">'svm_predict_sub'</span> LANGUAGE C IMMUTABLE STRICT; |
| <a name="l00431"></a>00431 |
| <a name="l00432"></a>00432 CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svm_predict(svs MADLIB_SCHEMA.svm_model_rec, ind float8[], kernel text) |
| <a name="l00433"></a>00433 RETURNS float8 AS $$ |
| <a name="l00434"></a>00434 SELECT MADLIB_SCHEMA.svm_predict_sub($1.nsvs, $1.ind_dim, $1.weights, $1.individuals, $2, $3); |
| <a name="l00435"></a>00435 $$ LANGUAGE SQL; |
| <a name="l00436"></a><a class="code" href="online__sv_8sql__in.html#a1ac76fdf9623e0a4db47665f2a80be90">00436</a> |
| <a name="l00437"></a>00437 -- This is the main online support vector regression learning algorithm. |
| <a name="l00438"></a>00438 -- The <span class="keyword">function</span> updates the support vector model as it processes each <span class="keyword">new</span> training example. |
| <a name="l00439"></a>00439 -- This <span class="keyword">function</span> is wrapped in an aggregate <span class="keyword">function</span> to process all the training examples stored in a table. |
| <a name="l00440"></a>00440 -- |
| <a name="l00441"></a>00441 CREATE OR REPLACE FUNCTION |
| <a name="l00442"></a>00442 MADLIB_SCHEMA.svm_reg_update(svs MADLIB_SCHEMA.svm_model_rec, ind FLOAT8[], label FLOAT8, kernel TEXT, eta FLOAT8, nu FLOAT8, slambda FLOAT8) |
| <a name="l00443"></a>00443 RETURNS MADLIB_SCHEMA.svm_model_rec AS <span class="stringliteral">'MODULE_PATHNAME'</span>, <span class="stringliteral">'svm_reg_update'</span> LANGUAGE C STRICT; |
| <a name="l00444"></a>00444 |
| <a name="l00445"></a>00445 CREATE AGGREGATE MADLIB_SCHEMA.svm_reg_agg(float8[], float8, text, float8, float8, float8) ( |
| <a name="l00446"></a>00446 sfunc = MADLIB_SCHEMA.svm_reg_update, |
| <a name="l00447"></a>00447 stype = MADLIB_SCHEMA.svm_model_rec, |
| <a name="l00448"></a><a class="code" href="online__sv_8sql__in.html#a9f2a96e1a241ecc66386a78b110777d3">00448</a> initcond = <span class="stringliteral">'(0,0,0,0,0,0,0,{},{},0)'</span> |
| <a name="l00449"></a>00449 ); |
| <a name="l00450"></a>00450 |
| <a name="l00451"></a>00451 -- This is the main online support vector classification learning algorithm. |
| <a name="l00452"></a>00452 -- The <span class="keyword">function</span> updates the support vector model as it processes each <span class="keyword">new</span> training example. |
| <a name="l00453"></a>00453 -- This <span class="keyword">function</span> is wrapped in an aggregate <span class="keyword">function</span> to process all the training examples stored in a table. |
| <a name="l00454"></a>00454 -- |
| <a name="l00455"></a>00455 CREATE OR REPLACE FUNCTION |
| <a name="l00456"></a>00456 MADLIB_SCHEMA.svm_cls_update(svs MADLIB_SCHEMA.svm_model_rec, ind FLOAT8[], label FLOAT8, kernel TEXT, eta FLOAT8, nu FLOAT8) |
| <a name="l00457"></a>00457 RETURNS MADLIB_SCHEMA.svm_model_rec AS <span class="stringliteral">'MODULE_PATHNAME'</span>, <span class="stringliteral">'svm_cls_update'</span> LANGUAGE C STRICT; |
| <a name="l00458"></a>00458 |
| <a name="l00459"></a>00459 CREATE AGGREGATE MADLIB_SCHEMA.svm_cls_agg(float8[], float8, text, float8, float8) ( |
| <a name="l00460"></a>00460 sfunc = MADLIB_SCHEMA.svm_cls_update, |
| <a name="l00461"></a>00461 stype = MADLIB_SCHEMA.svm_model_rec, |
| <a name="l00462"></a>00462 initcond = <span class="stringliteral">'(0,0,0,0,0,0,0,{},{},0)'</span> |
| <a name="l00463"></a>00463 ); |
| <a name="l00464"></a>00464 |
| <a name="l00465"></a>00465 -- This is the main online support vector novelty detection algorithm. |
| <a name="l00466"></a>00466 -- The <span class="keyword">function</span> updates the support vector model as it processes each <span class="keyword">new</span> training example. |
| <a name="l00467"></a>00467 -- In contrast to classification and regression, the training data points have no labels. |
| <a name="l00468"></a>00468 -- This <span class="keyword">function</span> is wrapped in an aggregate <span class="keyword">function</span> to process all the training examples stored in a table. |
| <a name="l00469"></a>00469 -- |
| <a name="l00470"></a>00470 CREATE OR REPLACE FUNCTION |
| <a name="l00471"></a>00471 MADLIB_SCHEMA.svm_nd_update(svs MADLIB_SCHEMA.svm_model_rec, ind FLOAT8[], kernel TEXT, eta FLOAT8, nu FLOAT8) |
| <a name="l00472"></a>00472 RETURNS MADLIB_SCHEMA.svm_model_rec AS <span class="stringliteral">'MODULE_PATHNAME'</span>, <span class="stringliteral">'svm_nd_update'</span> LANGUAGE C STRICT; |
| <a name="l00473"></a>00473 |
| <a name="l00474"></a>00474 CREATE AGGREGATE MADLIB_SCHEMA.svm_nd_agg(float8[], text, float8, float8) ( |
| <a name="l00475"></a>00475 sfunc = MADLIB_SCHEMA.svm_nd_update, |
| <a name="l00476"></a>00476 stype = MADLIB_SCHEMA.svm_model_rec, |
| <a name="l00477"></a>00477 initcond = <span class="stringliteral">'(0,0,0,0,0,0,0,{},{},0)'</span> |
| <a name="l00478"></a>00478 ); |
| <a name="l00479"></a>00479 |
| <a name="l00480"></a>00480 -- This is the SGD algorithm <span class="keywordflow">for</span> linear SVMs. |
| <a name="l00481"></a>00481 -- The <span class="keyword">function</span> updates the support vector model as it processes each <span class="keyword">new</span> training example. |
| <a name="l00482"></a>00482 -- This <span class="keyword">function</span> is wrapped in an aggregate <span class="keyword">function</span> to process all the training examples stored in a table. |
| <a name="l00483"></a>00483 -- |
| <a name="l00484"></a>00484 CREATE OR REPLACE FUNCTION |
| <a name="l00485"></a>00485 MADLIB_SCHEMA.lsvm_sgd_update(svs MADLIB_SCHEMA.lsvm_sgd_model_rec, ind FLOAT8[], label FLOAT8, eta FLOAT8, reg FLOAT8) |
| <a name="l00486"></a>00486 RETURNS MADLIB_SCHEMA.lsvm_sgd_model_rec AS <span class="stringliteral">'MODULE_PATHNAME'</span>, <span class="stringliteral">'lsvm_sgd_update'</span> LANGUAGE C STRICT; |
| <a name="l00487"></a>00487 |
| <a name="l00488"></a>00488 CREATE AGGREGATE MADLIB_SCHEMA.lsvm_sgd_agg(float8[], float8, float8, float8) ( |
| <a name="l00489"></a>00489 sfunc = MADLIB_SCHEMA.lsvm_sgd_update, |
| <a name="l00490"></a>00490 stype = MADLIB_SCHEMA.lsvm_sgd_model_rec, |
| <a name="l00491"></a>00491 initcond = <span class="stringliteral">'({},1,0,0,0,0)'</span> |
| <a name="l00492"></a>00492 ); |
| <a name="l00493"></a>00493 |
| <a name="l00494"></a>00494 |
| <a name="l00495"></a>00495 -- This <span class="keyword">function</span> stores a MADLIB_SCHEMA.svm_model_rec stored in model_temp_table into the model_table. |
| <a name="l00496"></a>00496 -- |
| <a name="l00497"></a>00497 CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svm_store_model(model_temp_table TEXT, model_name TEXT, model_table TEXT) RETURNS VOID AS $$ |
| <a name="l00498"></a>00498 |
| <a name="l00499"></a>00499 sql = <span class="stringliteral">"SELECT COUNT(*) FROM "</span> + model_temp_table + <span class="stringliteral">" WHERE id = \'"</span> + model_name + <span class="stringliteral">"\'"</span>; |
| <a name="l00500"></a>00500 temp = plpy.execute(sql); |
| <a name="l00501"></a>00501 <span class="keywordflow">if</span> (temp[0][<span class="stringliteral">'count'</span>] == 0): |
| <a name="l00502"></a>00502 plpy.error(<span class="stringliteral">"No support vector model with name "</span> + model_name + <span class="stringliteral">" found."</span>); |
| <a name="l00503"></a>00503 |
| <a name="l00504"></a>00504 sql = <span class="stringliteral">"SELECT (model).ind_dim, (model).nsvs"</span> \ |
| <a name="l00505"></a>00505 + <span class="stringliteral">" FROM "</span> + model_temp_table + <span class="stringliteral">" WHERE id = '"</span> + model_name + <span class="stringliteral">"'"</span>; |
| <a name="l00506"></a>00506 rv = plpy.execute(sql); |
| <a name="l00507"></a>00507 myind_dim = rv[0][<span class="stringliteral">'ind_dim'</span>]; |
| <a name="l00508"></a>00508 mynsvs = rv[0][<span class="stringliteral">'nsvs'</span>]; |
| <a name="l00509"></a>00509 |
| <a name="l00510"></a>00510 <span class="keywordflow">if</span> (mynsvs == 0): |
| <a name="l00511"></a>00511 plpy.error(<span class="stringliteral">"The specified model has no support vectors and therefore not processed"</span>); |
| <a name="l00512"></a>00512 |
| <a name="l00513"></a>00513 idx = 0; |
| <a name="l00514"></a>00514 <span class="keywordflow">for</span> i in range(1,mynsvs+1): |
| <a name="l00515"></a>00515 idx = myind_dim * (i-1); |
| <a name="l00516"></a>00516 sql = <span class="stringliteral">"INSERT INTO "</span> + model_table \ |
| <a name="l00517"></a>00517 + <span class="stringliteral">" SELECT \'"</span> + model_name + <span class="stringliteral">"\', (model).weights["</span> + str(i) + <span class="stringliteral">"], "</span> \ |
| <a name="l00518"></a>00518 + <span class="stringliteral">" (model).individuals[("</span> + str(idx+1) + <span class="stringliteral">"):("</span> + str(idx) + <span class="stringliteral">"+"</span> + str(myind_dim) + <span class="stringliteral">")] "</span> \ |
| <a name="l00519"></a>00519 + <span class="stringliteral">" FROM "</span> + model_temp_table + <span class="stringliteral">" WHERE id = \'"</span> + model_name + <span class="stringliteral">"\' LIMIT 1"</span>; |
| <a name="l00520"></a>00520 plpy.execute(sql); |
| <a name="l00521"></a>00521 |
| <a name="l00522"></a>00522 $$ LANGUAGE plpythonu; |
| <a name="l00523"></a>00523 <span class="comment"></span> |
| <a name="l00524"></a>00524 <span class="comment">/**</span> |
| <a name="l00525"></a>00525 <span class="comment"> * @brief Drops all tables pertaining to a model</span> |
| <a name="l00526"></a>00526 <span class="comment"> *</span> |
| <a name="l00527"></a>00527 <span class="comment"> * @param model_table The table to be dropped.</span> |
| <a name="l00528"></a>00528 <span class="comment"> */</span> |
| <a name="l00529"></a>00529 CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svm_drop_model(model_table TEXT) RETURNS VOID AS $$ |
| <a name="l00530"></a>00530 plpy.execute(<span class="stringliteral">"drop table if exists "</span> + model_table) |
| <a name="l00531"></a>00531 plpy.execute(<span class="stringliteral">"drop table if exists "</span> + model_table + <span class="stringliteral">"_param"</span>) |
| <a name="l00532"></a>00532 $$ LANGUAGE plpythonu; |
| <a name="l00533"></a>00533 |
| <a name="l00534"></a>00534 CREATE TYPE MADLIB_SCHEMA.svm_model_pr AS ( model text, prediction float8 ); |
| <a name="l00535"></a>00535 <span class="comment"></span> |
| <a name="l00536"></a>00536 <span class="comment">/**</span> |
| <a name="l00537"></a>00537 <span class="comment"> * @brief Evaluates a support-vector model on a given data point</span> |
| <a name="l00538"></a>00538 <span class="comment"> *</span> |
| <a name="l00539"></a>00539 <span class="comment"> * @param model_table The table storing the learned model \f$ f \f$ to be used</span> |
| <a name="l00540"></a>00540 <span class="comment"> * @param ind The data point \f$ \boldsymbol x \f$</span> |
| <a name="l00541"></a>00541 <span class="comment"> * @return This function returns \f$ f(\boldsymbol x) \f$</span> |
| <a name="l00542"></a>00542 <span class="comment"> */</span> |
| <a name="l00543"></a>00543 CREATE OR REPLACE FUNCTION |
| <a name="l00544"></a>00544 MADLIB_SCHEMA.svm_predict(model_table text, ind float8[]) RETURNS FLOAT8 AS $$ |
| <a name="l00545"></a>00545 |
| <a name="l00546"></a>00546 PythonFunctionBodyOnly(`kernel_machines<span class="stringliteral">', `online_sv'</span>) |
| <a name="l00547"></a>00547 |
| <a name="l00548"></a>00548 <span class="preprocessor"># schema_madlib comes from PythonFunctionBodyOnly</span> |
| <a name="l00549"></a>00549 <span class="preprocessor"></span> <span class="keywordflow">return</span> online_sv.svm_predict(model_table, ind); |
| <a name="l00550"></a>00550 |
| <a name="l00551"></a><a class="code" href="online__sv_8sql__in.html#ab54d33f13c0e00faa358e3e3f17c10fb">00551</a> $$ LANGUAGE plpythonu; |
| <a name="l00552"></a>00552 <span class="comment"></span> |
| <a name="l00553"></a>00553 <span class="comment">/**</span> |
| <a name="l00554"></a>00554 <span class="comment"> * @brief Evaluates multiple support-vector models on a data point</span> |
| <a name="l00555"></a>00555 <span class="comment"> *</span> |
| <a name="l00556"></a>00556 <span class="comment"> * @param model_table The table storing the learned models to be used.</span> |
| <a name="l00557"></a>00557 <span class="comment"> * @param ind The data point \f$ \boldsymbol x \f$</span> |
| <a name="l00558"></a>00558 <span class="comment"> * @return This function returns a table, a row for each model.</span> |
| <a name="l00559"></a>00559 <span class="comment"> * Moreover, the last row contains the average value, over all models.</span> |
| <a name="l00560"></a>00560 <span class="comment"> *</span> |
| <a name="l00561"></a>00561 <span class="comment"> * The different models are assumed to be named <tt><em>model_table</em>1</tt>,</span> |
| <a name="l00562"></a>00562 <span class="comment"> * <tt><em>model_table</em>2</tt>, ....</span> |
| <a name="l00563"></a>00563 <span class="comment"> */</span> |
| <a name="l00564"></a>00564 CREATE OR REPLACE FUNCTION |
| <a name="l00565"></a><a class="code" href="online__sv_8sql__in.html#a9916305653d464b23ef0fbd78867a654">00565</a> MADLIB_SCHEMA.svm_predict_combo(model_table text, ind float8[]) RETURNS SETOF MADLIB_SCHEMA.svm_model_pr AS $$ |
| <a name="l00566"></a>00566 |
| <a name="l00567"></a>00567 PythonFunctionBodyOnly(`kernel_machines<span class="stringliteral">', `online_sv'</span>) |
| <a name="l00568"></a>00568 |
| <a name="l00569"></a>00569 <span class="preprocessor"># schema_madlib comes from PythonFunctionBodyOnly</span> |
| <a name="l00570"></a>00570 <span class="preprocessor"></span> <span class="keywordflow">return</span> online_sv.svm_predict_combo( schema_madlib, model_table, ind); |
| <a name="l00571"></a>00571 |
| <a name="l00572"></a>00572 $$ LANGUAGE plpythonu; |
| <a name="l00573"></a>00573 |
| <a name="l00574"></a>00574 <span class="comment"></span> |
| <a name="l00575"></a>00575 <span class="comment">/**</span> |
| <a name="l00576"></a>00576 <span class="comment"> * @brief This is the support vector regression function</span> |
| <a name="l00577"></a>00577 <span class="comment"> *</span> |
| <a name="l00578"></a>00578 <span class="comment"> * @param input_table The name of the table/view with the training data</span> |
| <a name="l00579"></a>00579 <span class="comment"> * @param model_table The name of the table under which we want to store the learned model</span> |
| <a name="l00580"></a>00580 <span class="comment"> * @param parallel A flag indicating whether the system should learn multiple models in parallel</span> |
| <a name="l00581"></a>00581 <span class="comment"> * @param kernel_func Kernel function</span> |
| <a name="l00582"></a>00582 <span class="comment"> * @return A summary of the learning process</span> |
| <a name="l00583"></a>00583 <span class="comment"> *</span> |
| <a name="l00584"></a>00584 <span class="comment"> * @internal </span> |
| <a name="l00585"></a>00585 <span class="comment"> * @sa This function is a wrapper for online_sv::svm_regression().</span> |
| <a name="l00586"></a><a class="code" href="online__sv_8sql__in.html#a883ff4ca340d19a11204b461dd388276">00586</a> <span class="comment"> */</span> |
| <a name="l00587"></a>00587 CREATE OR REPLACE FUNCTION |
| <a name="l00588"></a>00588 MADLIB_SCHEMA.svm_regression(input_table text, model_table text, parallel <span class="keywordtype">bool</span>, kernel_func text) |
| <a name="l00589"></a>00589 RETURNS SETOF MADLIB_SCHEMA.svm_reg_result |
| <a name="l00590"></a>00590 AS $$ |
| <a name="l00591"></a>00591 |
| <a name="l00592"></a>00592 PythonFunctionBodyOnly(`kernel_machines<span class="stringliteral">', `online_sv'</span>) |
| <a name="l00593"></a>00593 |
| <a name="l00594"></a>00594 <span class="preprocessor"># schema_madlib comes from PythonFunctionBodyOnly</span> |
| <a name="l00595"></a>00595 <span class="preprocessor"></span> <span class="keywordflow">return</span> online_sv.svm_regression( schema_madlib, input_table, model_table, parallel, kernel_func); |
| <a name="l00596"></a>00596 |
| <a name="l00597"></a>00597 $$ LANGUAGE <span class="stringliteral">'plpythonu'</span>; |
| <a name="l00598"></a>00598 <span class="comment"></span> |
| <a name="l00599"></a>00599 <span class="comment">/**</span> |
| <a name="l00600"></a>00600 <span class="comment"> * @brief This is the support vector regression function</span> |
| <a name="l00601"></a>00601 <span class="comment"> *</span> |
| <a name="l00602"></a>00602 <span class="comment"> * @param input_table The name of the table/view with the training data</span> |
| <a name="l00603"></a>00603 <span class="comment"> * @param model_table The name of the table under which we want to store the learned model</span> |
| <a name="l00604"></a>00604 <span class="comment"> * @param parallel A flag indicating whether the system should learn multiple models in parallel</span> |
| <a name="l00605"></a>00605 <span class="comment"> * @param kernel_func Kernel function</span> |
| <a name="l00606"></a>00606 <span class="comment"> * @param verbose Verbosity of reporting</span> |
| <a name="l00607"></a>00607 <span class="comment"> * @param eta Learning rate in (0,1] </span> |
| <a name="l00608"></a>00608 <span class="comment"> * @param nu Compression parameter in (0,1] associated with the fraction of training data that will become support vectors </span> |
| <a name="l00609"></a><a class="code" href="online__sv_8sql__in.html#acaf1f4aa3eec5710de5c03e368a4b106">00609</a> <span class="comment"> * @param slambda Regularisation parameter</span> |
| <a name="l00610"></a>00610 <span class="comment"> * @return A summary of the learning process</span> |
| <a name="l00611"></a>00611 <span class="comment"> *</span> |
| <a name="l00612"></a>00612 <span class="comment"> * @internal </span> |
| <a name="l00613"></a>00613 <span class="comment"> * @sa This function is a wrapper for online_sv::svm_regression().</span> |
| <a name="l00614"></a>00614 <span class="comment"> */</span> |
| <a name="l00615"></a>00615 CREATE OR REPLACE FUNCTION |
| <a name="l00616"></a>00616 MADLIB_SCHEMA.svm_regression(input_table text, model_table text, parallel <span class="keywordtype">bool</span>, kernel_func text, verbose <span class="keywordtype">bool</span>, eta float8, nu float8, slambda float8) |
| <a name="l00617"></a>00617 RETURNS SETOF MADLIB_SCHEMA.svm_reg_result |
| <a name="l00618"></a>00618 AS $$ |
| <a name="l00619"></a>00619 |
| <a name="l00620"></a>00620 PythonFunctionBodyOnly(`kernel_machines<span class="stringliteral">', `online_sv'</span>) |
| <a name="l00621"></a>00621 |
| <a name="l00622"></a>00622 <span class="preprocessor"># schema_madlib comes from PythonFunctionBodyOnly</span> |
| <a name="l00623"></a>00623 <span class="preprocessor"></span> <span class="keywordflow">return</span> online_sv.svm_regression( schema_madlib, input_table, model_table, parallel, kernel_func, verbose, eta, nu, slambda); |
| <a name="l00624"></a>00624 |
| <a name="l00625"></a>00625 $$ LANGUAGE <span class="stringliteral">'plpythonu'</span>; |
| <a name="l00626"></a>00626 <span class="comment"></span> |
| <a name="l00627"></a>00627 <span class="comment">/**</span> |
| <a name="l00628"></a>00628 <span class="comment"> * @brief This is the support vector classification function</span> |
| <a name="l00629"></a>00629 <span class="comment"> *</span> |
| <a name="l00630"></a>00630 <span class="comment"> * @param input_table The name of the table/view with the training data</span> |
| <a name="l00631"></a>00631 <span class="comment"> * @param model_table The name of the table under which we want to store the learned model</span> |
| <a name="l00632"></a>00632 <span class="comment"> * @param parallel A flag indicating whether the system should learn multiple models in parallel</span> |
| <a name="l00633"></a>00633 <span class="comment"> * @param kernel_func Kernel function</span> |
| <a name="l00634"></a>00634 <span class="comment"> * @return A summary of the learning process</span> |
| <a name="l00635"></a>00635 <span class="comment"> *</span> |
| <a name="l00636"></a>00636 <span class="comment"> * @internal </span> |
| <a name="l00637"></a><a class="code" href="online__sv_8sql__in.html#ac5cb9c20d6620b155ac872576a056f2a">00637</a> <span class="comment"> * @sa This function is a wrapper for online_sv::svm_classification().</span> |
| <a name="l00638"></a>00638 <span class="comment"> */</span> |
| <a name="l00639"></a>00639 CREATE OR REPLACE FUNCTION |
| <a name="l00640"></a>00640 MADLIB_SCHEMA.svm_classification(input_table text, model_table text, parallel <span class="keywordtype">bool</span>, kernel_func text) |
| <a name="l00641"></a>00641 RETURNS SETOF MADLIB_SCHEMA.svm_cls_result |
| <a name="l00642"></a>00642 AS $$ |
| <a name="l00643"></a>00643 |
| <a name="l00644"></a>00644 PythonFunctionBodyOnly(`kernel_machines<span class="stringliteral">', `online_sv'</span>) |
| <a name="l00645"></a>00645 |
| <a name="l00646"></a>00646 <span class="preprocessor"># schema_madlib comes from PythonFunctionBodyOnly</span> |
| <a name="l00647"></a>00647 <span class="preprocessor"></span> <span class="keywordflow">return</span> online_sv.svm_classification( schema_madlib, input_table, model_table, parallel, kernel_func); |
| <a name="l00648"></a>00648 |
| <a name="l00649"></a>00649 $$ LANGUAGE <span class="stringliteral">'plpythonu'</span>; |
| <a name="l00650"></a>00650 <span class="comment"></span> |
| <a name="l00651"></a>00651 <span class="comment">/**</span> |
| <a name="l00652"></a>00652 <span class="comment"> * @brief This is the support vector classification function</span> |
| <a name="l00653"></a>00653 <span class="comment"> *</span> |
| <a name="l00654"></a>00654 <span class="comment"> * @param input_table The name of the table/view with the training data</span> |
| <a name="l00655"></a>00655 <span class="comment"> * @param model_table The name of the table under which we want to store the learned model</span> |
| <a name="l00656"></a>00656 <span class="comment"> * @param parallel A flag indicating whether the system should learn multiple models in parallel</span> |
| <a name="l00657"></a>00657 <span class="comment"> * @param kernel_func Kernel function</span> |
| <a name="l00658"></a>00658 <span class="comment"> * @param verbose Verbosity of reporting</span> |
| <a name="l00659"></a>00659 <span class="comment"> * @param eta Learning rate in (0,1]</span> |
| <a name="l00660"></a>00660 <span class="comment"> * @param nu Compression parameter in (0,1] associated with the fraction of training data that will become support vectors</span> |
| <a name="l00661"></a><a class="code" href="online__sv_8sql__in.html#a20a2c8a905be6e922885e23e9dab0a4c">00661</a> <span class="comment"> * @return A summary of the learning process</span> |
| <a name="l00662"></a>00662 <span class="comment"> *</span> |
| <a name="l00663"></a>00663 <span class="comment"> * @internal </span> |
| <a name="l00664"></a>00664 <span class="comment"> * @sa This function is a wrapper for online_sv::svm_classification().</span> |
| <a name="l00665"></a>00665 <span class="comment"> */</span> |
| <a name="l00666"></a>00666 CREATE OR REPLACE FUNCTION |
| <a name="l00667"></a>00667 MADLIB_SCHEMA.svm_classification(input_table text, model_table text, parallel <span class="keywordtype">bool</span>, kernel_func text, verbose <span class="keywordtype">bool</span>, eta float8, nu float8) |
| <a name="l00668"></a>00668 RETURNS SETOF MADLIB_SCHEMA.svm_cls_result |
| <a name="l00669"></a>00669 AS $$ |
| <a name="l00670"></a>00670 |
| <a name="l00671"></a>00671 PythonFunctionBodyOnly(`kernel_machines<span class="stringliteral">', `online_sv'</span>) |
| <a name="l00672"></a>00672 |
| <a name="l00673"></a>00673 <span class="preprocessor"># schema_madlib comes from PythonFunctionBodyOnly</span> |
| <a name="l00674"></a>00674 <span class="preprocessor"></span> <span class="keywordflow">return</span> online_sv.svm_classification( schema_madlib, input_table, model_table, parallel, kernel_func, verbose, eta, nu); |
| <a name="l00675"></a>00675 |
| <a name="l00676"></a>00676 $$ LANGUAGE <span class="stringliteral">'plpythonu'</span>; |
| <a name="l00677"></a>00677 <span class="comment"></span> |
| <a name="l00678"></a>00678 <span class="comment">/**</span> |
| <a name="l00679"></a>00679 <span class="comment"> * @brief This is the support vector novelty detection function.</span> |
| <a name="l00680"></a>00680 <span class="comment"> * </span> |
| <a name="l00681"></a>00681 <span class="comment"> * @param input_table The name of the table/view with the training data</span> |
| <a name="l00682"></a>00682 <span class="comment"> * @param model_table The name of the table under which we want to store the learned model</span> |
| <a name="l00683"></a>00683 <span class="comment"> * @param parallel A flag indicating whether the system should learn multiple models in parallel</span> |
| <a name="l00684"></a>00684 <span class="comment"> * @param kernel_func Kernel function</span> |
| <a name="l00685"></a>00685 <span class="comment"> * @return A summary of the learning process</span> |
| <a name="l00686"></a>00686 <span class="comment"> *</span> |
| <a name="l00687"></a>00687 <span class="comment"> * @internal </span> |
| <a name="l00688"></a><a class="code" href="online__sv_8sql__in.html#ad90b6bf3b807f22d37b0e2b1893262f0">00688</a> <span class="comment"> * @sa This function is a wrapper for online_sv::svm_novelty_detection().</span> |
| <a name="l00689"></a>00689 <span class="comment"> */</span> |
| <a name="l00690"></a>00690 CREATE OR REPLACE FUNCTION |
| <a name="l00691"></a>00691 MADLIB_SCHEMA.svm_novelty_detection(input_table text, model_table text, parallel <span class="keywordtype">bool</span>, kernel_func text) |
| <a name="l00692"></a>00692 RETURNS SETOF MADLIB_SCHEMA.svm_nd_result |
| <a name="l00693"></a>00693 AS $$ |
| <a name="l00694"></a>00694 |
| <a name="l00695"></a>00695 PythonFunctionBodyOnly(`kernel_machines<span class="stringliteral">', `online_sv'</span>) |
| <a name="l00696"></a>00696 |
| <a name="l00697"></a>00697 <span class="preprocessor"># schema_madlib comes from PythonFunctionBodyOnly</span> |
| <a name="l00698"></a>00698 <span class="preprocessor"></span> <span class="keywordflow">return</span> online_sv.svm_novelty_detection( schema_madlib, input_table, model_table, parallel, kernel_func); |
| <a name="l00699"></a>00699 |
| <a name="l00700"></a>00700 $$ LANGUAGE <span class="stringliteral">'plpythonu'</span>; |
| <a name="l00701"></a>00701 <span class="comment"></span> |
| <a name="l00702"></a>00702 <span class="comment">/**</span> |
| <a name="l00703"></a>00703 <span class="comment"> * @brief This is the support vector novelty detection function.</span> |
| <a name="l00704"></a>00704 <span class="comment"> * </span> |
| <a name="l00705"></a>00705 <span class="comment"> * @param input_table The name of the table/view with the training data</span> |
| <a name="l00706"></a>00706 <span class="comment"> * @param model_table The name of the table under which we want to store the learned model</span> |
| <a name="l00707"></a>00707 <span class="comment"> * @param parallel A flag indicating whether the system should learn multiple models in parallel</span> |
| <a name="l00708"></a>00708 <span class="comment"> * @param kernel_func Kernel function</span> |
| <a name="l00709"></a>00709 <span class="comment"> * @param verbose Verbosity of reporting</span> |
| <a name="l00710"></a>00710 <span class="comment"> * @param eta Learning rate in (0,1]</span> |
| <a name="l00711"></a>00711 <span class="comment"> * @param nu Compression parameter in (0,1] associated with the fraction of training data that will become support vectors</span> |
| <a name="l00712"></a><a class="code" href="online__sv_8sql__in.html#a3448ea62ab57fe4cf177f5fa6b5db7d3">00712</a> <span class="comment"> * @return A summary of the learning process</span> |
| <a name="l00713"></a>00713 <span class="comment"> *</span> |
| <a name="l00714"></a>00714 <span class="comment"> * @internal </span> |
| <a name="l00715"></a>00715 <span class="comment"> * @sa This function is a wrapper for online_sv::svm_novelty_detection().</span> |
| <a name="l00716"></a>00716 <span class="comment"> */</span> |
| <a name="l00717"></a>00717 CREATE OR REPLACE FUNCTION |
| <a name="l00718"></a>00718 MADLIB_SCHEMA.svm_novelty_detection(input_table text, model_table text, parallel <span class="keywordtype">bool</span>, kernel_func text, verbose <span class="keywordtype">bool</span>, eta float8, nu float8) |
| <a name="l00719"></a>00719 RETURNS SETOF MADLIB_SCHEMA.svm_nd_result |
| <a name="l00720"></a>00720 AS $$ |
| <a name="l00721"></a>00721 |
| <a name="l00722"></a>00722 PythonFunctionBodyOnly(`kernel_machines<span class="stringliteral">', `online_sv'</span>) |
| <a name="l00723"></a>00723 |
| <a name="l00724"></a>00724 <span class="preprocessor"># schema_madlib comes from PythonFunctionBodyOnly</span> |
| <a name="l00725"></a>00725 <span class="preprocessor"></span> <span class="keywordflow">return</span> online_sv.svm_novelty_detection( schema_madlib, input_table, model_table, parallel, kernel_func, verbose, eta, nu); |
| <a name="l00726"></a>00726 |
| <a name="l00727"></a>00727 $$ LANGUAGE <span class="stringliteral">'plpythonu'</span>; |
| <a name="l00728"></a>00728 |
| <a name="l00729"></a>00729 <span class="comment"></span> |
| <a name="l00730"></a>00730 <span class="comment">/**</span> |
| <a name="l00731"></a>00731 <span class="comment"> * @brief Scores the data points stored in a table using a learned support-vector model</span> |
| <a name="l00732"></a>00732 <span class="comment"> *</span> |
| <a name="l00733"></a>00733 <span class="comment"> * @param input_table Name of table/view containing the data points to be scored</span> |
| <a name="l00734"></a>00734 <span class="comment"> * @param data_col Name of column in input_table containing the data points</span> |
| <a name="l00735"></a>00735 <span class="comment"> * @param id_col Name of column in input_table containing the integer identifier of data points</span> |
| <a name="l00736"></a>00736 <span class="comment"> * @param model_table Name of table where the learned model to be used is stored</span> |
| <a name="l00737"></a>00737 <span class="comment"> * @param output_table Name of table to store the results </span> |
| <a name="l00738"></a>00738 <span class="comment"> * @param parallel A flag indicating whether the model to be used was learned in parallel</span> |
| <a name="l00739"></a><a class="code" href="online__sv_8sql__in.html#a5bae5335b51e448cd7fb9cb7a54b0bfa">00739</a> <span class="comment"> * @return Textual summary of the algorithm run</span> |
| <a name="l00740"></a>00740 <span class="comment"> *</span> |
| <a name="l00741"></a>00741 <span class="comment"> * @internal </span> |
| <a name="l00742"></a>00742 <span class="comment"> * @sa This function is a wrapper for online_sv::svm_predict_batch().</span> |
| <a name="l00743"></a>00743 <span class="comment"> */</span> |
| <a name="l00744"></a>00744 CREATE OR REPLACE FUNCTION |
| <a name="l00745"></a>00745 MADLIB_SCHEMA.svm_predict_batch(input_table text, data_col text, id_col text, model_table text, output_table text, parallel <span class="keywordtype">bool</span>) |
| <a name="l00746"></a>00746 RETURNS TEXT |
| <a name="l00747"></a>00747 AS $$ |
| <a name="l00748"></a>00748 |
| <a name="l00749"></a>00749 PythonFunctionBodyOnly(`kernel_machines<span class="stringliteral">', `online_sv'</span>) |
| <a name="l00750"></a>00750 |
| <a name="l00751"></a>00751 <span class="preprocessor"># schema_madlib comes from PythonFunctionBodyOnly</span> |
| <a name="l00752"></a>00752 <span class="preprocessor"></span> <span class="keywordflow">return</span> online_sv.svm_predict_batch( input_table, data_col, id_col, model_table, output_table, parallel); |
| <a name="l00753"></a>00753 |
| <a name="l00754"></a>00754 $$ LANGUAGE <span class="stringliteral">'plpythonu'</span>; |
| <a name="l00755"></a>00755 |
| <a name="l00756"></a>00756 -- Generate artificial training data |
| <a name="l00757"></a>00757 CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.__svm_random_ind(d INT) RETURNS float8[] AS $$ |
| <a name="l00758"></a>00758 DECLARE |
| <a name="l00759"></a>00759 ret float8[]; |
| <a name="l00760"></a>00760 BEGIN |
| <a name="l00761"></a>00761 FOR i IN 1..(d-1) LOOP |
| <a name="l00762"></a>00762 ret[i] = RANDOM() * 40 - 20; |
| <a name="l00763"></a>00763 END LOOP; |
| <a name="l00764"></a>00764 IF (RANDOM() > 0.5) THEN |
| <a name="l00765"></a>00765 ret[d] = 10; |
| <a name="l00766"></a><a class="code" href="online__sv_8sql__in.html#a91ac71354e9dec74e25339bf168c2e5b">00766</a> ELSE |
| <a name="l00767"></a>00767 ret[d] = -10; |
| <a name="l00768"></a>00768 END IF; |
| <a name="l00769"></a>00769 RETURN ret; |
| <a name="l00770"></a>00770 END |
| <a name="l00771"></a>00771 $$ LANGUAGE plpgsql; |
| <a name="l00772"></a>00772 |
| <a name="l00773"></a>00773 CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.__svm_random_ind2(d INT) RETURNS float8[] AS $$ |
| <a name="l00774"></a>00774 DECLARE |
| <a name="l00775"></a>00775 ret float8[]; |
| <a name="l00776"></a>00776 BEGIN |
| <a name="l00777"></a>00777 FOR i IN 1..d LOOP |
| <a name="l00778"></a>00778 ret[i] = RANDOM() * 5 + 10; |
| <a name="l00779"></a>00779 IF (RANDOM() > 0.5) THEN ret[i] = -ret[i]; END IF; |
| <a name="l00780"></a>00780 END LOOP; |
| <a name="l00781"></a>00781 RETURN ret; |
| <a name="l00782"></a>00782 END |
| <a name="l00783"></a>00783 $$ LANGUAGE plpgsql; |
| <a name="l00784"></a>00784 |
| <a name="l00785"></a>00785 |
| <a name="l00786"></a>00786 CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svm_generate_reg_data(output_table text, num <span class="keywordtype">int</span>, dim <span class="keywordtype">int</span>) RETURNS VOID AS $$ |
| <a name="l00787"></a>00787 plpy.execute("drop table if exists " + output_table) |
| <a name="l00788"></a>00788 plpy.execute("create table " + output_table + " ( <span class="keywordtype">id</span> <span class="keywordtype">int</span>, ind float8[], label float8 ) m4_ifdef(`__GREENPLUM__', `distributed by (<span class="keywordtype">id</span>)')") |
| <a name="l00789"></a>00789 plpy.execute("INSERT INTO " + output_table + " SELECT a.val, MADLIB_SCHEMA.__svm_random_ind(" + str(dim) + "), 0 FROM (SELECT generate_series(1," + str(num) + ") AS val) AS a") |
| <a name="l00790"></a>00790 plpy.execute("UPDATE " + output_table + " SET label = MADLIB_SCHEMA.__svm_target_reg_func(ind)") |
| <a name="l00791"></a>00791 $$ LANGUAGE 'plpythonu'; |
| <a name="l00792"></a>00792 |
| <a name="l00793"></a>00793 CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.__svm_target_reg_func(ind float8[]) RETURNS float8 AS $$ |
| <a name="l00794"></a>00794 DECLARE |
| <a name="l00795"></a>00795 dim <span class="keywordtype">int</span>; |
| <a name="l00796"></a>00796 BEGIN |
| <a name="l00797"></a>00797 dim = array_upper(ind,1); |
| <a name="l00798"></a>00798 IF (ind[dim] = 10) THEN RETURN 50; END IF; |
| <a name="l00799"></a>00799 RETURN -50; |
| <a name="l00800"></a>00800 END |
| <a name="l00801"></a>00801 $$ LANGUAGE plpgsql; |
| <a name="l00802"></a>00802 |
| <a name="l00803"></a>00803 CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svm_generate_cls_data(output_table text, num <span class="keywordtype">int</span>, dim <span class="keywordtype">int</span>) RETURNS VOID AS $$ |
| <a name="l00804"></a>00804 plpy.execute("drop table if exists " + output_table); |
| <a name="l00805"></a>00805 plpy.execute("create table " + output_table + " ( <span class="keywordtype">id</span> <span class="keywordtype">int</span>, ind float8[], label float8 ) m4_ifdef(`__GREENPLUM__', `distributed by (<span class="keywordtype">id</span>)')") |
| <a name="l00806"></a>00806 plpy.execute("INSERT INTO " + output_table + " SELECT a.val, MADLIB_SCHEMA.__svm_random_ind(" + str(dim) + "), 0 FROM (SELECT generate_series(1," + str(num) + ") AS val) AS a") |
| <a name="l00807"></a>00807 plpy.execute("UPDATE " + output_table + " SET label = MADLIB_SCHEMA.__svm_target_cl_func(ind)") |
| <a name="l00808"></a>00808 $$ LANGUAGE 'plpythonu'; |
| <a name="l00809"></a>00809 |
| <a name="l00810"></a>00810 CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.__svm_target_cl_func(ind float8[]) RETURNS float8 AS $$ |
| <a name="l00811"></a>00811 BEGIN |
| <a name="l00812"></a>00812 IF (ind[1] > 0 AND ind[2] < 0) THEN RETURN 1; END IF; |
| <a name="l00813"></a>00813 RETURN -1; |
| <a name="l00814"></a>00814 END |
| <a name="l00815"></a>00815 $$ LANGUAGE plpgsql; |
| <a name="l00816"></a>00816 |
| <a name="l00817"></a>00817 CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.svm_generate_nd_data(output_table text, num <span class="keywordtype">int</span>, dim <span class="keywordtype">int</span>) RETURNS VOID AS $$ |
| <a name="l00818"></a>00818 plpy.execute("drop table if exists " + output_table); |
| <a name="l00819"></a>00819 plpy.execute("create table " + output_table + " ( <span class="keywordtype">id</span> <span class="keywordtype">int</span>, ind float8[] ) m4_ifdef(`__GREENPLUM__', `distributed by (<span class="keywordtype">id</span>)')") |
| <a name="l00820"></a>00820 plpy.execute("INSERT INTO " + output_table + " SELECT a.val, MADLIB_SCHEMA.__svm_random_ind2(" + str(dim) + ") FROM (SELECT generate_series(1," + str(num) + ") AS val) AS a") |
| <a name="l00821"></a>00821 $$ LANGUAGE 'plpythonu'; |
| <a name="l00822"></a>00822 |
| <a name="l00823"></a>00823 <span class="comment"></span> |
| <a name="l00824"></a>00824 <span class="comment">/**</span> |
| <a name="l00825"></a>00825 <span class="comment"> * @brief Normalizes the data stored in a table, and save the normalized data in a new table. </span> |
| <a name="l00826"></a>00826 <span class="comment"> *</span> |
| <a name="l00827"></a>00827 <span class="comment"> * @param input_table Name of table/view containing the data points to be scored</span> |
| <a name="l00828"></a>00828 <span class="comment"> */</span> |
| <a name="l00829"></a>00829 CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.<a class="code" href="online__sv_8sql__in.html#a78bb07d8f4aee80c3bbc8e8176e512be" title="Normalizes the data stored in a table, and save the normalized data in a new table.">svm_data_normalization</a>(input_table TEXT) RETURNS VOID AS $$ |
| <a name="l00830"></a>00830 output_table = input_table + "_scaled"; |
| <a name="l00831"></a>00831 plpy.execute("DROP TABLE IF EXISTS " + output_table); |
| <a name="l00832"></a>00832 plpy.execute("CREATE TABLE " + output_table + " ( <span class="keywordtype">id</span> <span class="keywordtype">int</span>, ind float8[], label <span class="keywordtype">int</span> ) m4_ifdef(`__GREENPLUM__', `distributed by (<span class="keywordtype">id</span>)')"); |
| <a name="l00833"></a>00833 plpy.execute("INSERT INTO " + output_table + " SELECT <span class="keywordtype">id</span>, MADLIB_SCHEMA.svm_normalization(ind), label FROM " + input_table); |
| <a name="l00834"></a>00834 plpy.info("output table: %s" % output_table) |
| <a name="l00835"></a>00835 $$ LANGUAGE plpythonu; |
| <a name="l00836"></a>00836 |
| <a name="l00837"></a>00837 <span class="comment"></span> |
| <a name="l00838"></a>00838 <span class="comment">/**</span> |
| <a name="l00839"></a>00839 <span class="comment"> * @brief This is the linear support vector classification function</span> |
| <a name="l00840"></a>00840 <span class="comment"> *</span> |
| <a name="l00841"></a>00841 <span class="comment"> * @param input_table The name of the table/view with the training data</span> |
| <a name="l00842"></a>00842 <span class="comment"> * @param model_table The name of the table under which we want to store the learned model</span> |
| <a name="l00843"></a>00843 <span class="comment"> * @param parallel A flag indicating whether the system should learn multiple models in parallel</span> |
| <a name="l00844"></a>00844 <span class="comment"> * @return A summary of the learning process</span> |
| <a name="l00845"></a>00845 <span class="comment"> *</span> |
| <a name="l00846"></a>00846 <span class="comment"> * @internal </span> |
| <a name="l00847"></a>00847 <span class="comment"> * @sa This function is a wrapper for online_sv::lsvm_classification().</span> |
| <a name="l00848"></a>00848 <span class="comment">*/</span> |
| <a name="l00849"></a>00849 CREATE OR REPLACE FUNCTION |
| <a name="l00850"></a>00850 MADLIB_SCHEMA.<a class="code" href="online__sv_8sql__in.html#a75d126981ae4bf2e6641627501f0a2a5" title="This is the linear support vector classification function.">lsvm_classification</a>(input_table text, model_table text, parallel <span class="keywordtype">bool</span>) |
| <a name="l00851"></a><a class="code" href="online__sv_8sql__in.html#a78bb07d8f4aee80c3bbc8e8176e512be">00851</a> RETURNS SETOF MADLIB_SCHEMA.lsvm_sgd_result |
| <a name="l00852"></a>00852 AS $$ |
| <a name="l00853"></a>00853 PythonFunctionBodyOnly(`kernel_machines', `online_sv') |
| <a name="l00854"></a>00854 <span class="preprocessor"># schema_madlib comes from PythonFunctionBodyOnly</span> |
| <a name="l00855"></a>00855 <span class="preprocessor"></span> <span class="keywordflow">return</span> online_sv.lsvm_classification( schema_madlib, input_table, model_table, parallel); |
| <a name="l00856"></a>00856 $$ LANGUAGE <span class="stringliteral">'plpythonu'</span>; |
| <a name="l00857"></a>00857 |
| <a name="l00858"></a>00858 |
| <a name="l00859"></a>00859 <span class="comment"></span> |
| <a name="l00860"></a>00860 <span class="comment">/**</span> |
| <a name="l00861"></a>00861 <span class="comment"> * @brief This is the linear support vector classification function</span> |
| <a name="l00862"></a>00862 <span class="comment"> *</span> |
| <a name="l00863"></a>00863 <span class="comment"> * @param input_table The name of the table/view with the training data</span> |
| <a name="l00864"></a>00864 <span class="comment"> * @param model_table The name of the table under which we want to store the learned model</span> |
| <a name="l00865"></a>00865 <span class="comment"> * @param parallel A flag indicating whether the system should learn multiple models in parallel</span> |
| <a name="l00866"></a>00866 <span class="comment"> * @param verbose Verbosity of reporting</span> |
| <a name="l00867"></a>00867 <span class="comment"> * @param eta Initial learning rate in (0,1]</span> |
| <a name="l00868"></a>00868 <span class="comment"> * @param reg Regularization parameter, often chosen by cross-validation</span> |
| <a name="l00869"></a>00869 <span class="comment"> * @return A summary of the learning process</span> |
| <a name="l00870"></a>00870 <span class="comment"> *</span> |
| <a name="l00871"></a><a class="code" href="online__sv_8sql__in.html#a75d126981ae4bf2e6641627501f0a2a5">00871</a> <span class="comment"> * @internal </span> |
| <a name="l00872"></a>00872 <span class="comment"> * @sa This function is a wrapper for online_sv::lsvm_classification().</span> |
| <a name="l00873"></a>00873 <span class="comment">*/</span> |
| <a name="l00874"></a>00874 CREATE OR REPLACE FUNCTION |
| <a name="l00875"></a>00875 MADLIB_SCHEMA.lsvm_classification(input_table text, model_table text, parallel <span class="keywordtype">bool</span>, verbose <span class="keywordtype">bool</span>, eta float8, reg float8) |
| <a name="l00876"></a>00876 RETURNS SETOF MADLIB_SCHEMA.lsvm_sgd_result |
| <a name="l00877"></a>00877 AS $$ |
| <a name="l00878"></a>00878 |
| <a name="l00879"></a>00879 PythonFunctionBodyOnly(`kernel_machines<span class="stringliteral">', `online_sv'</span>) |
| <a name="l00880"></a>00880 |
| <a name="l00881"></a>00881 <span class="preprocessor"># schema_madlib comes from PythonFunctionBodyOnly</span> |
| <a name="l00882"></a>00882 <span class="preprocessor"></span> <span class="keywordflow">return</span> online_sv.lsvm_classification( schema_madlib, input_table, model_table, parallel, verbose, eta, reg); |
| <a name="l00883"></a>00883 |
| <a name="l00884"></a>00884 $$ LANGUAGE <span class="stringliteral">'plpythonu'</span>; |
| <a name="l00885"></a>00885 |
| <a name="l00886"></a>00886 <span class="comment"></span> |
| <a name="l00887"></a>00887 <span class="comment">/**</span> |
| <a name="l00888"></a>00888 <span class="comment"> * @brief Scores the data points stored in a table using a learned linear support-vector model</span> |
| <a name="l00889"></a>00889 <span class="comment"> *</span> |
| <a name="l00890"></a>00890 <span class="comment"> * @param input_table Name of table/view containing the data points to be scored</span> |
| <a name="l00891"></a>00891 <span class="comment"> * @param data_col Name of column in input_table containing the data points</span> |
| <a name="l00892"></a>00892 <span class="comment"> * @param id_col Name of column in input_table containing the integer identifier of data points</span> |
| <a name="l00893"></a>00893 <span class="comment"> * @param model_table Name of table where the learned model to be used is stored</span> |
| <a name="l00894"></a>00894 <span class="comment"> * @param output_table Name of table to store the results </span> |
| <a name="l00895"></a>00895 <span class="comment"> * @param parallel A flag indicating whether the model to be used was learned in parallel</span> |
| <a name="l00896"></a><a class="code" href="online__sv_8sql__in.html#a50896def00d0e0950bec3d95b387e6b9">00896</a> <span class="comment"> * @return Textual summary of the algorithm run</span> |
| <a name="l00897"></a>00897 <span class="comment"> *</span> |
| <a name="l00898"></a>00898 <span class="comment"> * @internal </span> |
| <a name="l00899"></a>00899 <span class="comment"> * @sa This function is a wrapper for online_sv::lsvm_predict_batch().</span> |
| <a name="l00900"></a>00900 <span class="comment"> */</span> |
| <a name="l00901"></a>00901 CREATE OR REPLACE FUNCTION |
| <a name="l00902"></a>00902 MADLIB_SCHEMA.lsvm_predict_batch(input_table text, data_col text, id_col text, model_table text, output_table text, parallel <span class="keywordtype">bool</span>) |
| <a name="l00903"></a>00903 RETURNS TEXT |
| <a name="l00904"></a>00904 AS $$ |
| <a name="l00905"></a>00905 |
| <a name="l00906"></a>00906 PythonFunctionBodyOnly(`kernel_machines<span class="stringliteral">', `online_sv'</span>) |
| <a name="l00907"></a>00907 |
| <a name="l00908"></a>00908 <span class="preprocessor"># schema_madlib comes from PythonFunctionBodyOnly</span> |
| <a name="l00909"></a>00909 <span class="preprocessor"></span> <span class="keywordflow">return</span> online_sv.lsvm_predict_batch( schema_madlib, input_table, data_col, id_col, model_table, output_table, parallel); |
| <a name="l00910"></a>00910 |
| <a name="l00911"></a>00911 $$ LANGUAGE <span class="stringliteral">'plpythonu'</span>; |
| <a name="l00912"></a>00912 |
| <a name="l00913"></a>00913 <span class="comment"></span> |
| <a name="l00914"></a>00914 <span class="comment">/**</span> |
| <a name="l00915"></a>00915 <span class="comment"> * @brief Evaluates a linear support-vector model on a given data point</span> |
| <a name="l00916"></a>00916 <span class="comment"> *</span> |
| <a name="l00917"></a>00917 <span class="comment"> * @param model_table The table storing the learned model \f$ f \f$ to be used</span> |
| <a name="l00918"></a>00918 <span class="comment"> * @param ind The data point \f$ \boldsymbol x \f$</span> |
| <a name="l00919"></a>00919 <span class="comment"> * @return This function returns \f$ f(\boldsymbol x) \f$</span> |
| <a name="l00920"></a>00920 <span class="comment"> */</span> |
| <a name="l00921"></a>00921 CREATE OR REPLACE FUNCTION |
| <a name="l00922"></a>00922 MADLIB_SCHEMA.lsvm_predict(model_table text, ind float8[]) RETURNS FLOAT8 AS $$ |
| <a name="l00923"></a><a class="code" href="online__sv_8sql__in.html#a1c0a002f50250133c0ef1d3c43c6d338">00923</a> |
| <a name="l00924"></a>00924 PythonFunctionBodyOnly(`kernel_machines<span class="stringliteral">', `online_sv'</span>) |
| <a name="l00925"></a>00925 |
| <a name="l00926"></a>00926 <span class="preprocessor"># schema_madlib comes from PythonFunctionBodyOnly</span> |
| <a name="l00927"></a>00927 <span class="preprocessor"></span> <span class="keywordflow">return</span> online_sv.lsvm_predict(schema_madlib, model_table, ind); |
| <a name="l00928"></a>00928 |
| <a name="l00929"></a>00929 $$ LANGUAGE plpythonu; |
| <a name="l00930"></a>00930 <span class="comment"></span> |
| <a name="l00931"></a>00931 <span class="comment">/**</span> |
| <a name="l00932"></a>00932 <span class="comment"> * @brief Evaluates multiple linear support-vector models on a data point</span> |
| <a name="l00933"></a>00933 <span class="comment"> *</span> |
| <a name="l00934"></a>00934 <span class="comment"> * @param model_table The table storing the learned models to be used.</span> |
| <a name="l00935"></a>00935 <span class="comment"> * @param ind The data point \f$ \boldsymbol x \f$</span> |
| <a name="l00936"></a>00936 <span class="comment"> * @return This function returns a table, a row for each model.</span> |
| <a name="l00937"></a>00937 <span class="comment"> * Moreover, the last row contains the average value, over all models.</span> |
| <a name="l00938"></a>00938 <span class="comment"> *</span> |
| <a name="l00939"></a>00939 <span class="comment"> * The different models are assumed to be named <tt><em>model_table</em>0</tt>,</span> |
| <a name="l00940"></a>00940 <span class="comment"> * <tt><em>model_table</em>1</tt>, ....</span> |
| <a name="l00941"></a>00941 <span class="comment"> */</span> |
| <a name="l00942"></a>00942 CREATE OR REPLACE FUNCTION |
| <a name="l00943"></a><a class="code" href="online__sv_8sql__in.html#a5fe084c8364c0657097410458f8ea1e9">00943</a> MADLIB_SCHEMA.lsvm_predict_combo(model_table text, ind float8[]) RETURNS SETOF MADLIB_SCHEMA.svm_model_pr AS $$ |
| <a name="l00944"></a>00944 |
| <a name="l00945"></a>00945 PythonFunctionBodyOnly(`kernel_machines<span class="stringliteral">', `online_sv'</span>) |
| <a name="l00946"></a>00946 |
| <a name="l00947"></a>00947 <span class="preprocessor"># schema_madlib comes from PythonFunctionBodyOnly</span> |
| <a name="l00948"></a>00948 <span class="preprocessor"></span> <span class="keywordflow">return</span> online_sv.lsvm_predict_combo( schema_madlib, model_table, ind); |
| <a name="l00949"></a>00949 |
| <a name="l00950"></a>00950 $$ LANGUAGE plpythonu; |
| </pre></div></div> |
| </div> |
| <div id="nav-path" class="navpath"> |
| <ul> |
| <li class="navelem"><a class="el" href="online__sv_8sql__in.html">online_sv.sql_in</a> </li> |
| <!-- window showing the filter options --> |
| <div id="MSearchSelectWindow" |
| onmouseover="return searchBox.OnSearchSelectShow()" |
| onmouseout="return searchBox.OnSearchSelectHide()" |
| onkeydown="return searchBox.OnSearchSelectKey(event)"> |
| <a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(0)"><span class="SelectionMark"> </span>All</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(1)"><span class="SelectionMark"> </span>Files</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(2)"><span class="SelectionMark"> </span>Functions</a></div> |
| |
| <!-- iframe showing the search results (closed by default) --> |
| <div id="MSearchResultsWindow"> |
| <iframe src="javascript:void(0)" frameborder="0" |
| name="MSearchResults" id="MSearchResults"> |
| </iframe> |
| </div> |
| |
| |
| <li class="footer">Generated on Tue Apr 2 2013 14:57:03 for MADlib by |
| <a href="http://www.doxygen.org/index.html"> |
| <img class="footer" src="doxygen.png" alt="doxygen"/></a> 1.7.5.1 </li> |
| </ul> |
| </div> |
| |
| |
| </body> |
| </html> |
