http://git-wip-us.apache.org/repos/asf/madlib-site/blob/af0e5f14/docs/v1.15.1/group__grp__mfvsketch.html ---------------------------------------------------------------------- diff --git a/docs/v1.15.1/group__grp__mfvsketch.html b/docs/v1.15.1/group__grp__mfvsketch.html new file mode 100644 index 0000000..aa76e83 --- /dev/null +++ b/docs/v1.15.1/group__grp__mfvsketch.html @@ -0,0 +1,183 @@ +<!-- HTML header for doxygen 1.8.4--> +<!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"/> +<meta http-equiv="X-UA-Compatible" content="IE=9"/> +<meta name="generator" content="Doxygen 1.8.14"/> +<meta name="keywords" content="madlib,postgres,greenplum,machine learning,data mining,deep learning,ensemble methods,data science,market basket analysis,affinity analysis,pca,lda,regression,elastic net,huber white,proportional hazards,k-means,latent dirichlet allocation,bayes,support vector machines,svm"/> +<title>MADlib: MFV (Most Frequent Values)</title> +<link href="tabs.css" rel="stylesheet" type="text/css"/> +<script type="text/javascript" src="jquery.js"></script> +<script type="text/javascript" src="dynsections.js"></script> +<link href="navtree.css" rel="stylesheet" type="text/css"/> +<script type="text/javascript" src="resize.js"></script> +<script type="text/javascript" src="navtreedata.js"></script> +<script type="text/javascript" src="navtree.js"></script> +<script type="text/javascript"> +/* @license magnet:?xt=urn:btih:cf05388f2679ee054f2beb29a391d25f4e673ac3&dn=gpl-2.0.txt GPL-v2 */ + $(document).ready(initResizable); +/* @license-end */</script> +<link href="search/search.css" rel="stylesheet" type="text/css"/> +<script type="text/javascript" src="search/searchdata.js"></script> +<script type="text/javascript" src="search/search.js"></script> +<script type="text/javascript"> +/* @license magnet:?xt=urn:btih:cf05388f2679ee054f2beb29a391d25f4e673ac3&dn=gpl-2.0.txt GPL-v2 */ + $(document).ready(function() { init_search(); }); +/* @license-end */ +</script> +<script type="text/x-mathjax-config"> + MathJax.Hub.Config({ + extensions: ["tex2jax.js", "TeX/AMSmath.js", "TeX/AMSsymbols.js"], + jax: ["input/TeX","output/HTML-CSS"], +}); +</script><script type="text/javascript" async src="https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/MathJax.js";></script> +<!-- hack in the navigation tree --> +<script type="text/javascript" src="eigen_navtree_hacks.js"></script> +<link href="doxygen.css" rel="stylesheet" type="text/css" /> +<link href="madlib_extra.css" rel="stylesheet" type="text/css"/> +<!-- google analytics --> +<script> + (function(i,s,o,g,r,a,m){i['GoogleAnalyticsObject']=r;i[r]=i[r]||function(){ + (i[r].q=i[r].q||[]).push(arguments)},i[r].l=1*new Date();a=s.createElement(o), + m=s.getElementsByTagName(o)[0];a.async=1;a.src=g;m.parentNode.insertBefore(a,m) + })(window,document,'script','//www.google-analytics.com/analytics.js','ga'); + ga('create', 'UA-45382226-1', 'madlib.apache.org'); + ga('send', 'pageview'); +</script> +</head> +<body> +<div id="top"><!-- do not remove this div, it is closed by doxygen! --> +<div id="titlearea"> +<table cellspacing="0" cellpadding="0"> + <tbody> + <tr style="height: 56px;"> + <td id="projectlogo"><a href="http://madlib.apache.org";><img alt="Logo" src="madlib.png" height="50" style="padding-left:0.5em;" border="0"/ ></a></td> + <td style="padding-left: 0.5em;"> + <div id="projectname"> + <span id="projectnumber">1.15.1</span> + </div> + <div id="projectbrief">User Documentation for Apache MADlib</div> + </td> + <td> <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> +</td> + </tr> + </tbody> +</table> +</div> +<!-- end header part --> +<!-- Generated by Doxygen 1.8.14 --> +<script type="text/javascript"> +/* @license magnet:?xt=urn:btih:cf05388f2679ee054f2beb29a391d25f4e673ac3&dn=gpl-2.0.txt GPL-v2 */ +var searchBox = new SearchBox("searchBox", "search",false,'Search'); +/* @license-end */ +</script> +</div><!-- top --> +<div id="side-nav" class="ui-resizable side-nav-resizable"> + <div id="nav-tree"> + <div id="nav-tree-contents"> + <div id="nav-sync" class="sync"></div> + </div> + </div> + <div id="splitbar" style="-moz-user-select:none;" + class="ui-resizable-handle"> + </div> +</div> +<script type="text/javascript"> +/* @license magnet:?xt=urn:btih:cf05388f2679ee054f2beb29a391d25f4e673ac3&dn=gpl-2.0.txt GPL-v2 */ +$(document).ready(function(){initNavTree('group__grp__mfvsketch.html','');}); +/* @license-end */ +</script> +<div id="doc-content"> +<!-- window showing the filter options --> +<div id="MSearchSelectWindow" + onmouseover="return searchBox.OnSearchSelectShow()" + onmouseout="return searchBox.OnSearchSelectHide()" + onkeydown="return searchBox.OnSearchSelectKey(event)"> +</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> + +<div class="header"> + <div class="headertitle"> +<div class="title">MFV (Most Frequent Values)<div class="ingroups"><a class="el" href="group__grp__stats.html">Statistics</a> » <a class="el" href="group__grp__desc__stats.html">Descriptive Statistics</a> » <a class="el" href="group__grp__sketches.html">Cardinality Estimators</a></div></div> </div> +</div><!--header--> +<div class="contents"> +<div class="toc"><b>Contents</b> <ul> +<li> +<a href="#syntax">Syntax</a> </li> +<li> +<a href="#examples">Examples</a> </li> +<li> +<a href="#literature">Literature</a> </li> +<li> +<a href="#related">Related Topics</a> </li> +</ul> +</div><p>MFVSketch: Most Frequent Values variant of CountMin sketch, implemented as a UDA.</p> +<p>Produces an n-bucket histogram for a column where each bucket counts one of the most frequent values in the column. The output is an array of doubles {value, count} in descending order of frequency; counts are approximated via CountMin sketches. Ties are handled arbitrarily.</p> +<p><a class="anchor" id="syntax"></a> The MFV frequent-value UDA comes in two different versions:</p><ul> +<li>a faithful implementation that preserves the approximation guarantees of Cormode/Muthukrishnan, <pre class="syntax"> +mfvsketch_top_histogram( col_name, + n ) +</pre></li> +<li>and a "quick and dirty" version that can do parallel aggregation in Greenplum at the expense of missing some of the most frequent values. <pre class="syntax"> +mfvsketch_quick_histogram( col_name, + n ) +</pre></li> +</ul> +<p>In PostgreSQL the two UDAs are identical. In Greenplum, the quick version should produce good results unless the number of values requested is small, or the distribution is flat.</p> +<dl class="section note"><dt>Note</dt><dd>This is a <a href="https://www.postgresql.org/docs/current/static/xaggr.html";>User Defined Aggregate</a> which returns the results when used in a query. Use "CREATE TABLE AS ", with the UDA as subquery if the results are to be stored. This is unlike the usual MADlib stored procedure interface which places the results in a table instead of returning it.</dd></dl> +<p><a class="anchor" id="examples"></a></p><dl class="section user"><dt>Examples</dt><dd></dd></dl> +<ol type="1"> +<li>Generate some data. <pre class="example"> +CREATE TABLE data(class INT, a1 INT); +INSERT INTO data SELECT 1,1 FROM generate_series(1,10000); +INSERT INTO data SELECT 1,2 FROM generate_series(1,15000); +INSERT INTO data SELECT 1,3 FROM generate_series(1,10000); +INSERT INTO data SELECT 2,5 FROM generate_series(1,1000); +INSERT INTO data SELECT 2,6 FROM generate_series(1,1000); +</pre></li> +<li>Produce a histogram of 5 bins and return the most frequent value and associated count in each bin. <pre class="example"> +SELECT mfvsketch_top_histogram( a1, 5 ) +FROM data; +</pre> Result: <pre class="result"> + mfvsketch_top_histogram + ------------------------------------------------------------- +[0:4][0:1]={{2,15000},{1,10000},{3,10000},{5,1000},{6,1000}} +(1 row) +</pre></li> +</ol> +<p><a class="anchor" id="literature"></a></p><dl class="section user"><dt>Literature</dt><dd>This method is not usually called an MFV sketch in the literature; it is a natural extension of the CountMin sketch.</dd></dl> +<p><a class="anchor" id="related"></a></p><dl class="section user"><dt>Related Topics</dt><dd></dd></dl> +<p>File <a class="el" href="sketch_8sql__in.html" title="SQL functions for sketch-based approximations of descriptive statistics. ">sketch.sql_in</a> documenting the SQL functions.</p> +<p>Module <a class="el" href="group__grp__countmin.html">CountMin (Cormode-Muthukrishnan)</a>. </p> +</div><!-- contents --> +</div><!-- doc-content --> +<!-- start footer part --> +<div id="nav-path" class="navpath"><!-- id is needed for treeview function! --> + <ul> + <li class="footer">Generated on Mon Oct 15 2018 11:24:30 for MADlib by + <a href="http://www.doxygen.org/index.html";> + <img class="footer" src="doxygen.png" alt="doxygen"/></a> 1.8.14 </li> + </ul> +</div> +</body> +</html>
http://git-wip-us.apache.org/repos/asf/madlib-site/blob/af0e5f14/docs/v1.15.1/group__grp__minibatch__preprocessing.html ---------------------------------------------------------------------- diff --git a/docs/v1.15.1/group__grp__minibatch__preprocessing.html b/docs/v1.15.1/group__grp__minibatch__preprocessing.html new file mode 100644 index 0000000..0fa58d5 --- /dev/null +++ b/docs/v1.15.1/group__grp__minibatch__preprocessing.html @@ -0,0 +1,462 @@ +<!-- HTML header for doxygen 1.8.4--> +<!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"/> +<meta http-equiv="X-UA-Compatible" content="IE=9"/> +<meta name="generator" content="Doxygen 1.8.14"/> +<meta name="keywords" content="madlib,postgres,greenplum,machine learning,data mining,deep learning,ensemble methods,data science,market basket analysis,affinity analysis,pca,lda,regression,elastic net,huber white,proportional hazards,k-means,latent dirichlet allocation,bayes,support vector machines,svm"/> +<title>MADlib: Mini-Batch Preprocessor</title> +<link href="tabs.css" rel="stylesheet" type="text/css"/> +<script type="text/javascript" src="jquery.js"></script> +<script type="text/javascript" src="dynsections.js"></script> +<link href="navtree.css" rel="stylesheet" type="text/css"/> +<script type="text/javascript" src="resize.js"></script> +<script type="text/javascript" src="navtreedata.js"></script> +<script type="text/javascript" src="navtree.js"></script> +<script type="text/javascript"> +/* @license magnet:?xt=urn:btih:cf05388f2679ee054f2beb29a391d25f4e673ac3&dn=gpl-2.0.txt GPL-v2 */ + $(document).ready(initResizable); +/* @license-end */</script> +<link href="search/search.css" rel="stylesheet" type="text/css"/> +<script type="text/javascript" src="search/searchdata.js"></script> +<script type="text/javascript" src="search/search.js"></script> +<script type="text/javascript"> +/* @license magnet:?xt=urn:btih:cf05388f2679ee054f2beb29a391d25f4e673ac3&dn=gpl-2.0.txt GPL-v2 */ + $(document).ready(function() { init_search(); }); +/* @license-end */ +</script> +<script type="text/x-mathjax-config"> + MathJax.Hub.Config({ + extensions: ["tex2jax.js", "TeX/AMSmath.js", "TeX/AMSsymbols.js"], + jax: ["input/TeX","output/HTML-CSS"], +}); +</script><script type="text/javascript" async src="https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/MathJax.js";></script> +<!-- hack in the navigation tree --> +<script type="text/javascript" src="eigen_navtree_hacks.js"></script> +<link href="doxygen.css" rel="stylesheet" type="text/css" /> +<link href="madlib_extra.css" rel="stylesheet" type="text/css"/> +<!-- google analytics --> +<script> + (function(i,s,o,g,r,a,m){i['GoogleAnalyticsObject']=r;i[r]=i[r]||function(){ + (i[r].q=i[r].q||[]).push(arguments)},i[r].l=1*new Date();a=s.createElement(o), + m=s.getElementsByTagName(o)[0];a.async=1;a.src=g;m.parentNode.insertBefore(a,m) + })(window,document,'script','//www.google-analytics.com/analytics.js','ga'); + ga('create', 'UA-45382226-1', 'madlib.apache.org'); + ga('send', 'pageview'); +</script> +</head> +<body> +<div id="top"><!-- do not remove this div, it is closed by doxygen! --> +<div id="titlearea"> +<table cellspacing="0" cellpadding="0"> + <tbody> + <tr style="height: 56px;"> + <td id="projectlogo"><a href="http://madlib.apache.org";><img alt="Logo" src="madlib.png" height="50" style="padding-left:0.5em;" border="0"/ ></a></td> + <td style="padding-left: 0.5em;"> + <div id="projectname"> + <span id="projectnumber">1.15.1</span> + </div> + <div id="projectbrief">User Documentation for Apache MADlib</div> + </td> + <td> <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> +</td> + </tr> + </tbody> +</table> +</div> +<!-- end header part --> +<!-- Generated by Doxygen 1.8.14 --> +<script type="text/javascript"> +/* @license magnet:?xt=urn:btih:cf05388f2679ee054f2beb29a391d25f4e673ac3&dn=gpl-2.0.txt GPL-v2 */ +var searchBox = new SearchBox("searchBox", "search",false,'Search'); +/* @license-end */ +</script> +</div><!-- top --> +<div id="side-nav" class="ui-resizable side-nav-resizable"> + <div id="nav-tree"> + <div id="nav-tree-contents"> + <div id="nav-sync" class="sync"></div> + </div> + </div> + <div id="splitbar" style="-moz-user-select:none;" + class="ui-resizable-handle"> + </div> +</div> +<script type="text/javascript"> +/* @license magnet:?xt=urn:btih:cf05388f2679ee054f2beb29a391d25f4e673ac3&dn=gpl-2.0.txt GPL-v2 */ +$(document).ready(function(){initNavTree('group__grp__minibatch__preprocessing.html','');}); +/* @license-end */ +</script> +<div id="doc-content"> +<!-- window showing the filter options --> +<div id="MSearchSelectWindow" + onmouseover="return searchBox.OnSearchSelectShow()" + onmouseout="return searchBox.OnSearchSelectHide()" + onkeydown="return searchBox.OnSearchSelectKey(event)"> +</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> + +<div class="header"> + <div class="headertitle"> +<div class="title">Mini-Batch Preprocessor<div class="ingroups"><a class="el" href="group__grp__other__functions.html">Utilities</a></div></div> </div> +</div><!--header--> +<div class="contents"> +<div class="toc"><b>Contents</b><ul> +<li class="level1"> +<a href="#minibatch_preprocessor">Mini-Batch Preprocessor</a> </li> +<li class="level1"> +<a href="#example">Examples</a> </li> +<li class="level1"> +<a href="#literature">Literature</a> </li> +<li class="level1"> +<a href="#related">Related Topics</a> </li> +</ul> +</div><p>The mini-batch preprocessor is a utility that prepares input data for use by models that support mini-batch as an optimization option. (This is currently only the case for <a href="group__grp__nn.html">Neural Networks</a>.) It is effectively a packing operation that builds arrays of dependent and independent variables from the source data table.</p> +<p>The advantage of using mini-batching is that it can perform better than stochastic gradient descent (default MADlib optimizer) because it uses more than one training example at a time, typically resulting faster and smoother convergence [1].</p> +<p><a class="anchor" id="minibatch_preprocessor"></a></p><dl class="section user"><dt>Mini-Batch Preprocessor</dt><dd>The mini-batch preprocessor has the following format:</dd></dl> +<pre class="syntax"> +minibatch_preprocessor( source_table, + output_table, + dependent_varname, + independent_varname, + grouping_col, + buffer_size, + one_hot_encode_int_dep_var + ) +</pre><p><b>Arguments</b> </p><dl class="arglist"> +<dt>source_table </dt> +<dd><p class="startdd">TEXT. Name of the table containing input data. Can also be a view. </p> +<p class="enddd"></p> +</dd> +<dt>output_table </dt> +<dd><p class="startdd">TEXT. Name of the output table from the preprocessor which will be used as input to algorithms that support mini-batching. Note that the arrays packed into the output table are randomized and normalized, so they will not match up in an obvious way with the rows in the source table. </p> +<p class="enddd"></p> +</dd> +<dt>dependent_varname </dt> +<dd><p class="startdd">TEXT. Name of the dependent variable column. </p> +<p class="enddd"></p> +</dd> +<dt>independent_varname </dt> +<dd>TEXT. Column name or expression list to evaluate for the independent variable. Please note that independent variables are cast to double precision by the preprocessor, so categorical variables should be one-hot or dummy encoded as appropriate. See <a href="group__grp__encode__categorical.html">Encoding Categorical Variables</a> for more details on this. <dl class="section note"><dt>Note</dt><dd>Supported expressions for independent variables include:<ul> +<li>âARRAY[x1,x2,x3]â, where x1, x2, and x3 are columns in the source table containing scalar values.</li> +<li>Single column in the source table containing an array like ARRAY[1,2,3] or {1,2,3}. </li> +</ul> +</dd> +<dd> +The following forms are not currently supported:<ul> +<li>âx1,x2,x3â, where x1,x2,x3 are columns in source table with scalar values</li> +<li>ARRAY[x1,x2] where x1 is scalar and x2 is array</li> +<li>ARRAY[x1,x2] where both x1 and x2 are arrays</li> +<li>ARRAY[x1] where x1 is array </li> +</ul> +</dd></dl> +</dd> +<dt>grouping_col (optional) </dt> +<dd>TEXT, default: NULL. An expression list used to group the input dataset into discrete groups, which runs the preprocessing separately for each group. When this value is NULL, no grouping is used and a single preprocessor step is run for the whole data set. <dl class="section note"><dt>Note</dt><dd>If you plan to use grouping in model training, then you must set up the groups in the preprocessor exactly as you want to use them in training. </dd></dl> +</dd> +<dt>buffer_size (optional) </dt> +<dd><p class="startdd">INTEGER, default: computed. Buffer size is the number of rows from the source table that are packed into one row of the preprocessor output table. The default value is computed considering size of the source table, number of independent variables, number of groups, and number of segments in the database cluster. For larger data sets, the computed buffer size will typically be a value in the millions. </p> +<p class="enddd"></p> +</dd> +<dt>one_hot_encode_int_dep_var (optional) </dt> +<dd><p class="startdd">BOOLEAN. default: FALSE. Flag to one-hot encode dependent variables that are scalar integers. This parameter is ignored if the dependent variable is not a scalar integer.</p> +<dl class="section note"><dt>Note</dt><dd>The mini-batch preprocessor automatically encodes dependent variables that are boolean and character types such as text, char and varchar. However, scalar integers are a special case because they can be used in both classification and regression problems, so you must tell the mini-batch preprocessor whether you want to encode them or not. In the case that you have already encoded the dependent variable yourself, you can ignore this parameter. Also, if you want to encode float values for some reason, cast them to text first. </dd></dl> +</dd> +</dl> +<p><b>Output tables</b> <br /> + The output table produced by the mini-batch preprocessor contains the following columns: </p><table class="output"> +<tr> +<th>__id__ </th><td>INTEGER. Unique id for packed table. </td></tr> +<tr> +<th>dependent_varname </th><td>FLOAT8[]. Packed array of dependent variables. If the dependent variable in the source table is categorical, the preprocessor will one-hot encode it. </td></tr> +<tr> +<th>independent_varname </th><td>FLOAT8[]. Packed array of independent variables. </td></tr> +<tr> +<th>grouping_cols </th><td>TEXT. Name of grouping columns. </td></tr> +</table> +<p>A summary table named <output_table>_summary is also created, which has the following columns: </p><table class="output"> +<tr> +<th>source_table </th><td>Name of the source table. </td></tr> +<tr> +<th>output_table </th><td>Name of output table generated by preprocessor. </td></tr> +<tr> +<th>dependent_varname </th><td>Dependent variable from the source table. </td></tr> +<tr> +<th>independent_varname </th><td>Independent variable from the source table. </td></tr> +<tr> +<th>buffer_size </th><td>Buffer size used in preprocessing step. </td></tr> +<tr> +<th>class_values </th><td>Class values (i.e., levels) of the dependent variable if categorical. If the dependent variable is not categorical, this will be NULL./td> </td></tr> +<tr> +<th>num_rows_processed </th><td>The total number of rows that were used in the preprocessing operation. </td></tr> +<tr> +<th>num_missing_rows_skipped </th><td>The total number of rows that were skipped because of NULL values in either the dependent or independent variables. </td></tr> +<tr> +<th>grouping_col </th><td>Comma separated list of grouping column names if grouping is used. If no grouping, will be NULL. </td></tr> +</table> +<p>A standardization table named <output_table>_standardization is also created. This is needed by the models that will use the preprocessed data so is likely not of much interest to users. It has the following columns: </p><table class="output"> +<tr> +<th>grouping columns </th><td>If 'grouping_col' is specified, a column for each grouping column is created. </td></tr> +<tr> +<th>mean </th><td>Mean of independent variables. </td></tr> +<tr> +<th>std </th><td>Population standard deviation of independent variables. </td></tr> +</table> +<p><a class="anchor" id="example"></a></p><dl class="section user"><dt>Examples</dt><dd><ol type="1"> +<li>Create an input data set based on the well known iris data set: <pre class="example"> +DROP TABLE IF EXISTS iris_data; +CREATE TABLE iris_data( + id serial, + attributes numeric[], + class_text varchar, + class integer, + state varchar +); +INSERT INTO iris_data(id, attributes, class_text, class, state) VALUES +(1,ARRAY[5.0,3.2,1.2,0.2],'Iris_setosa',1,'Alaska'), +(2,ARRAY[5.5,3.5,1.3,0.2],'Iris_setosa',1,'Alaska'), +(3,ARRAY[4.9,3.1,1.5,0.1],'Iris_setosa',1,'Alaska'), +(4,ARRAY[4.4,3.0,1.3,0.2],'Iris_setosa',1,'Alaska'), +(5,ARRAY[5.1,3.4,1.5,0.2],'Iris_setosa',1,'Alaska'), +(6,ARRAY[5.0,3.5,1.3,0.3],'Iris_setosa',1,'Alaska'), +(7,ARRAY[4.5,2.3,1.3,0.3],'Iris_setosa',1,'Alaska'), +(8,ARRAY[4.4,3.2,1.3,0.2],'Iris_setosa',1,'Alaska'), +(9,ARRAY[5.0,3.5,1.6,0.6],'Iris_setosa',1,'Alaska'), +(10,ARRAY[5.1,3.8,1.9,0.4],'Iris_setosa',1,'Alaska'), +(11,ARRAY[4.8,3.0,1.4,0.3],'Iris_setosa',1,'Alaska'), +(12,ARRAY[5.1,3.8,1.6,0.2],'Iris_setosa',1,'Alaska'), +(13,ARRAY[5.7,2.8,4.5,1.3],'Iris_versicolor',2,'Alaska'), +(14,ARRAY[6.3,3.3,4.7,1.6],'Iris_versicolor',2,'Alaska'), +(15,ARRAY[4.9,2.4,3.3,1.0],'Iris_versicolor',2,'Alaska'), +(16,ARRAY[6.6,2.9,4.6,1.3],'Iris_versicolor',2,'Alaska'), +(17,ARRAY[5.2,2.7,3.9,1.4],'Iris_versicolor',2,'Alaska'), +(18,ARRAY[5.0,2.0,3.5,1.0],'Iris_versicolor',2,'Alaska'), +(19,ARRAY[5.9,3.0,4.2,1.5],'Iris_versicolor',2,'Alaska'), +(20,ARRAY[6.0,2.2,4.0,1.0],'Iris_versicolor',2,'Alaska'), +(21,ARRAY[6.1,2.9,4.7,1.4],'Iris_versicolor',2,'Alaska'), +(22,ARRAY[5.6,2.9,3.6,1.3],'Iris_versicolor',2,'Alaska'), +(23,ARRAY[6.7,3.1,4.4,1.4],'Iris_versicolor',2,'Alaska'), +(24,ARRAY[5.6,3.0,4.5,1.5],'Iris_versicolor',2,'Alaska'), +(25,ARRAY[5.8,2.7,4.1,1.0],'Iris_versicolor',2,'Alaska'), +(26,ARRAY[6.2,2.2,4.5,1.5],'Iris_versicolor',2,'Alaska'), +(27,ARRAY[5.6,2.5,3.9,1.1],'Iris_versicolor',2,'Alaska'), +(28,ARRAY[5.0,3.4,1.5,0.2],'Iris_setosa',1,'Tennessee'), +(29,ARRAY[4.4,2.9,1.4,0.2],'Iris_setosa',1,'Tennessee'), +(30,ARRAY[4.9,3.1,1.5,0.1],'Iris_setosa',1,'Tennessee'), +(31,ARRAY[5.4,3.7,1.5,0.2],'Iris_setosa',1,'Tennessee'), +(32,ARRAY[4.8,3.4,1.6,0.2],'Iris_setosa',1,'Tennessee'), +(33,ARRAY[4.8,3.0,1.4,0.1],'Iris_setosa',1,'Tennessee'), +(34,ARRAY[4.3,3.0,1.1,0.1],'Iris_setosa',1,'Tennessee'), +(35,ARRAY[5.8,4.0,1.2,0.2],'Iris_setosa',1,'Tennessee'), +(36,ARRAY[5.7,4.4,1.5,0.4],'Iris_setosa',1,'Tennessee'), +(37,ARRAY[5.4,3.9,1.3,0.4],'Iris_setosa',1,'Tennessee'), +(38,ARRAY[6.0,2.9,4.5,1.5],'Iris_versicolor',2,'Tennessee'), +(39,ARRAY[5.7,2.6,3.5,1.0],'Iris_versicolor',2,'Tennessee'), +(40,ARRAY[5.5,2.4,3.8,1.1],'Iris_versicolor',2,'Tennessee'), +(41,ARRAY[5.5,2.4,3.7,1.0],'Iris_versicolor',2,'Tennessee'), +(42,ARRAY[5.8,2.7,3.9,1.2],'Iris_versicolor',2,'Tennessee'), +(43,ARRAY[6.0,2.7,5.1,1.6],'Iris_versicolor',2,'Tennessee'), +(44,ARRAY[5.4,3.0,4.5,1.5],'Iris_versicolor',2,'Tennessee'), +(45,ARRAY[6.0,3.4,4.5,1.6],'Iris_versicolor',2,'Tennessee'), +(46,ARRAY[6.7,3.1,4.7,1.5],'Iris_versicolor',2,'Tennessee'), +(47,ARRAY[6.3,2.3,4.4,1.3],'Iris_versicolor',2,'Tennessee'), +(48,ARRAY[5.6,3.0,4.1,1.3],'Iris_versicolor',2,'Tennessee'), +(49,ARRAY[5.5,2.5,4.0,1.3],'Iris_versicolor',2,'Tennessee'), +(50,ARRAY[5.5,2.6,4.4,1.2],'Iris_versicolor',2,'Tennessee'), +(51,ARRAY[6.1,3.0,4.6,1.4],'Iris_versicolor',2,'Tennessee'), +(52,ARRAY[5.8,2.6,4.0,1.2],'Iris_versicolor',2,'Tennessee'); +</pre></li> +<li>Run the preprocessor: <pre class="example"> +DROP TABLE IF EXISTS iris_data_packed, iris_data_packed_summary, iris_data_packed_standardization; +SELECT madlib.minibatch_preprocessor('iris_data', -- Source table + 'iris_data_packed', -- Output table + 'class_text', -- Dependent variable + 'attributes' -- Independent variables + ); +</pre> For small datasets like in this example, buffer size is mainly determined by the number of segments in the database. This example is run on a Greenplum database with 2 segments, so there are 2 rows with a buffer size of 26. For PostgresSQL, there would be only one row with a buffer size of 52 since it is a single node database. For larger data sets, other factors go into computing buffers size besides number of segments. Also, note that the dependent variable has been one-hot encoded since it is categorical. Here is a sample of the packed output table: <pre class="example"> +\x on +SELECT * FROM iris_data_packed; +</pre> <pre class="result"> +-[ RECORD 1 ]-------+------------------------------------- +__id__ | 0 +dependent_varname | {{1,0},{0,1},{1,0},{0,1},{1,0},{0,1},{0,1},{1,0},{1,0},{1,0},{1,0},{0,1},{0,1},{0,1},{1,0},{0,1},{0,1},{0,1},{1,0},{0,1},{1,0},{0,1},{1,0},{1,0},{1,0},{0,1}} +independent_varname | {{-0.767560815504508,0.806649237861967,-1.07515071152907,-1.18456909732025},{-0.0995580974152422,0.00385956572525086,1.03989986852812,1.17758048907675},... +... +-[ RECORD 2 ]-------+------------------------------------- +__id__ | 1 +dependent_varname | {{1,0},{1,0},{1,0},{0,1},{0,1},{1,0},{0,1},{0,1},{0,1},{0,1},{0,1},{0,1},{0,1},{1,0},{0,1},{0,1},{0,1},{0,1},{0,1},{1,0},{0,1},{1,0},{0,1},{1,0},{1,0},{0,1}} +independent_varname | {{0.568444620674023,2.01083374606704,-1.28665576953479,-1.18456909732025},{-1.76956489263841,0.405254401793609,-1.21615408353289,-1.18456909732025},... +... +</pre> Review the output summary table: <pre class="example"> +SELECT * FROM iris_data_packed_summary; +</pre> <pre class="result"> +-[ RECORD 1 ]------------+------------------------------ +source_table | iris_data +output_table | iris_data_packed +dependent_varname | class_text +independent_varname | attributes +buffer_size | 26 +class_values | {Iris_setosa,Iris_versicolor} +num_rows_processed | 52 +num_missing_rows_skipped | 0 +grouping_cols | +</pre> Review the output standardization table: <pre class="example"> +SELECT * FROM iris_data_packed_standardization; +</pre> <pre class="result"> +-[ RECORD 1 ]------------------------------------------------------ +mean | {5.45961538462,2.99807692308,3.025,0.851923076923} +std | {0.598799958695,0.498262513686,1.41840579525,0.550346179381} +</pre></li> +<li>Generally the default buffer size will work well, but if you have occasion to change it: <pre class="example"> +DROP TABLE IF EXISTS iris_data_packed, iris_data_packed_summary, iris_data_packed_standardization; +SELECT madlib.minibatch_preprocessor('iris_data', -- Source table + 'iris_data_packed', -- Output table + 'class_text', -- Dependent variable + 'attributes', -- Independent variables + NULL, -- Grouping + 10 -- Buffer size + ); +</pre> Review the output summary table: <pre class="example"> +SELECT * FROM iris_data_packed_summary; +</pre> <pre class="result"> +-[ RECORD 1 ]------------+------------------------------ +source_table | iris_data +output_table | iris_data_packed +dependent_varname | class_text +independent_varname | attributes +buffer_size | 10 +class_values | {Iris_setosa,Iris_versicolor} +num_rows_processed | 52 +num_missing_rows_skipped | 0 +grouping_cols | +</pre></li> +<li>Run the preprocessor with grouping by state: <pre class="example"> +DROP TABLE IF EXISTS iris_data_packed, iris_data_packed_summary, iris_data_packed_standardization; +SELECT madlib.minibatch_preprocessor('iris_data', -- Source table + 'iris_data_packed', -- Output table + 'class_text', -- Dependent variable + 'attributes', -- Independent variables + 'state' -- Grouping + ); +</pre> Review the output table: <pre class="example"> +SELECT * FROM iris_data_packed ORDER BY state, __id__; +</pre> <pre class="result"> +-[ RECORD 1 ]-------+------------------------------------- +__id__ | 0 +state | Alaska +dependent_varname | {{0,1},{0,1},{1,0},{0,1},{0,1},{0,1},{1,0},{0,1},{0,1},{1,0},{1,0},{0,1},{0,1}} +independent_varname | {{0.306242850830503,-0.977074857057813,0.680489757142278 ... +... +-[ RECORD 2 ]-------+------------------------------------- +__id__ | 1 +state | Alaska +dependent_varname | {{0,1},{1,0},{0,1},{0,1},{1,0},{1,0},{1,0},{0,1},{1,0},{0,1},{0,1},{1,0},{1,0}} +independent_varname | {{1.10129640587123,-0.126074175104234,1.2524188915498 ... +... +-[ RECORD 3 ]-------+------------------------------------- +__id__ | 2 +state | Alaska +dependent_varname | {{1,0}} +independent_varname | {{-0.647821415218373,1.15042684782613,-1.17827992968215 ... +... +-[ RECORD 4 ]-------+------------------------------------- +__id__ | 0 +state | Tennessee +dependent_varname | {{1,0},{0,1},{1,0},{1,0},{1,0},{0,1},{1,0},{0,1},{0,1},{0,1},{1,0},{1,0},{0,1}} +independent_varname | {{0.32912603663053,2.59625206429212,-1.12079945083087 ... +... +-[ RECORD 5 ]-------+------------------------------------- +__id__ | 1 +state | Tennessee +dependent_varname | {{0,1},{0,1},{0,1},{1,0},{1,0},{0,1},{0,1},{1,0},{0,1},{0,1},{0,1},{0,1}} +independent_varname | {{0.865744574615085,-0.267261241912424,0.970244300719264 ... +... +</pre> Review the output summary table: <pre class="example"> +SELECT * FROM iris_data_packed_summary; +</pre> <pre class="result"> +-[ RECORD 1 ]------------+------------------------------ +source_table | iris_data +output_table | iris_data_packed +dependent_varname | class_text +independent_varname | attributes +buffer_size | 13 +class_values | {Iris_setosa,Iris_versicolor} +num_rows_processed | 52 +num_missing_rows_skipped | 0 +grouping_cols | state +</pre> Review the output standardization table: <pre class="example"> +SELECT * FROM iris_data_packed_standardization; +</pre> <pre class="result"> +-[ RECORD 1 ]------------------------------------------------------------------- +state | Alaska +mean | {5.40740740740741,2.95925925925926,2.94814814814815,0.833333333333333} +std | {0.628888452645665,0.470034875978888,1.39877469405147,0.536103914747325} +-[ RECORD 2 ]------------------------------------------------------------------- +state | Tennessee +mean | {5.516,3.04,3.108,0.872} +std | {0.55905634778617,0.523832034148353,1.43469021046357,0.564637937088893} +</pre></li> +<li>If the depedent variable is scalar integer, and you have not already encoded it, you can ask the preprocessor to encode it for you: <pre class="example"> +DROP TABLE IF EXISTS iris_data_packed, iris_data_packed_summary, iris_data_packed_standardization; +SELECT madlib.minibatch_preprocessor('iris_data', -- Source table + 'iris_data_packed', -- Output table + 'class', -- Integer dependent variable + 'attributes', -- Independent variables + NULL, -- Grouping + NULL, -- Buffer size + TRUE -- Encode scalar int dependent variable + ); +</pre> Review the output summary table: <pre class="example"> +SELECT * FROM iris_data_packed_summary; +</pre> <pre class="result"> +-[ RECORD 1 ]------------+----------------- +source_table | iris_data +output_table | iris_data_packed +dependent_varname | class +independent_varname | attributes +dependent_vartype | integer +buffer_size | 26 +class_values | {1,2} +num_rows_processed | 52 +num_missing_rows_skipped | 0 +grouping_cols | +</pre></li> +</ol> +</dd></dl> +<p><a class="anchor" id="literature"></a></p><dl class="section user"><dt>Literature</dt><dd></dd></dl> +<p>[1] "Neural Networks for Machine Learning", Lectures 6a and 6b on mini-batch gradient descent, Geoffrey Hinton with Nitish Srivastava and Kevin Swersky, <a href="http://www.cs.toronto.edu/~tijmen/csc321/slides/lecture_slides_lec6.pdf";>http://www.cs.toronto.edu/~tijmen/csc321/slides/lecture_slides_lec6.pdf</a></p> +<p><a class="anchor" id="related"></a></p><dl class="section user"><dt>Related Topics</dt><dd></dd></dl> +<p><a class="el" href="minibatch__preprocessing_8sql__in.html" title="TODO. ">minibatch_preprocessing.sql_in</a></p> +<p><a href="group__grp__nn.html"><b>Neural Networks</b></a> </p> +</div><!-- contents --> +</div><!-- doc-content --> +<!-- start footer part --> +<div id="nav-path" class="navpath"><!-- id is needed for treeview function! --> + <ul> + <li class="footer">Generated on Mon Oct 15 2018 11:24:30 for MADlib by + <a href="http://www.doxygen.org/index.html";> + <img class="footer" src="doxygen.png" alt="doxygen"/></a> 1.8.14 </li> + </ul> +</div> +</body> +</html> http://git-wip-us.apache.org/repos/asf/madlib-site/blob/af0e5f14/docs/v1.15.1/group__grp__mlogreg.html ---------------------------------------------------------------------- diff --git a/docs/v1.15.1/group__grp__mlogreg.html b/docs/v1.15.1/group__grp__mlogreg.html new file mode 100644 index 0000000..7dbbd4d --- /dev/null +++ b/docs/v1.15.1/group__grp__mlogreg.html @@ -0,0 +1,423 @@ +<!-- HTML header for doxygen 1.8.4--> +<!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"/> +<meta http-equiv="X-UA-Compatible" content="IE=9"/> +<meta name="generator" content="Doxygen 1.8.14"/> +<meta name="keywords" content="madlib,postgres,greenplum,machine learning,data mining,deep learning,ensemble methods,data science,market basket analysis,affinity analysis,pca,lda,regression,elastic net,huber white,proportional hazards,k-means,latent dirichlet allocation,bayes,support vector machines,svm"/> +<title>MADlib: Multinomial Logistic Regression</title> +<link href="tabs.css" rel="stylesheet" type="text/css"/> +<script type="text/javascript" src="jquery.js"></script> +<script type="text/javascript" src="dynsections.js"></script> +<link href="navtree.css" rel="stylesheet" type="text/css"/> +<script type="text/javascript" src="resize.js"></script> +<script type="text/javascript" src="navtreedata.js"></script> +<script type="text/javascript" src="navtree.js"></script> +<script type="text/javascript"> +/* @license magnet:?xt=urn:btih:cf05388f2679ee054f2beb29a391d25f4e673ac3&dn=gpl-2.0.txt GPL-v2 */ + $(document).ready(initResizable); +/* @license-end */</script> +<link href="search/search.css" rel="stylesheet" type="text/css"/> +<script type="text/javascript" src="search/searchdata.js"></script> +<script type="text/javascript" src="search/search.js"></script> +<script type="text/javascript"> +/* @license magnet:?xt=urn:btih:cf05388f2679ee054f2beb29a391d25f4e673ac3&dn=gpl-2.0.txt GPL-v2 */ + $(document).ready(function() { init_search(); }); +/* @license-end */ +</script> +<script type="text/x-mathjax-config"> + MathJax.Hub.Config({ + extensions: ["tex2jax.js", "TeX/AMSmath.js", "TeX/AMSsymbols.js"], + jax: ["input/TeX","output/HTML-CSS"], +}); +</script><script type="text/javascript" async src="https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/MathJax.js";></script> +<!-- hack in the navigation tree --> +<script type="text/javascript" src="eigen_navtree_hacks.js"></script> +<link href="doxygen.css" rel="stylesheet" type="text/css" /> +<link href="madlib_extra.css" rel="stylesheet" type="text/css"/> +<!-- google analytics --> +<script> + (function(i,s,o,g,r,a,m){i['GoogleAnalyticsObject']=r;i[r]=i[r]||function(){ + (i[r].q=i[r].q||[]).push(arguments)},i[r].l=1*new Date();a=s.createElement(o), + m=s.getElementsByTagName(o)[0];a.async=1;a.src=g;m.parentNode.insertBefore(a,m) + })(window,document,'script','//www.google-analytics.com/analytics.js','ga'); + ga('create', 'UA-45382226-1', 'madlib.apache.org'); + ga('send', 'pageview'); +</script> +</head> +<body> +<div id="top"><!-- do not remove this div, it is closed by doxygen! --> +<div id="titlearea"> +<table cellspacing="0" cellpadding="0"> + <tbody> + <tr style="height: 56px;"> + <td id="projectlogo"><a href="http://madlib.apache.org";><img alt="Logo" src="madlib.png" height="50" style="padding-left:0.5em;" border="0"/ ></a></td> + <td style="padding-left: 0.5em;"> + <div id="projectname"> + <span id="projectnumber">1.15.1</span> + </div> + <div id="projectbrief">User Documentation for Apache MADlib</div> + </td> + <td> <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> +</td> + </tr> + </tbody> +</table> +</div> +<!-- end header part --> +<!-- Generated by Doxygen 1.8.14 --> +<script type="text/javascript"> +/* @license magnet:?xt=urn:btih:cf05388f2679ee054f2beb29a391d25f4e673ac3&dn=gpl-2.0.txt GPL-v2 */ +var searchBox = new SearchBox("searchBox", "search",false,'Search'); +/* @license-end */ +</script> +</div><!-- top --> +<div id="side-nav" class="ui-resizable side-nav-resizable"> + <div id="nav-tree"> + <div id="nav-tree-contents"> + <div id="nav-sync" class="sync"></div> + </div> + </div> + <div id="splitbar" style="-moz-user-select:none;" + class="ui-resizable-handle"> + </div> +</div> +<script type="text/javascript"> +/* @license magnet:?xt=urn:btih:cf05388f2679ee054f2beb29a391d25f4e673ac3&dn=gpl-2.0.txt GPL-v2 */ +$(document).ready(function(){initNavTree('group__grp__mlogreg.html','');}); +/* @license-end */ +</script> +<div id="doc-content"> +<!-- window showing the filter options --> +<div id="MSearchSelectWindow" + onmouseover="return searchBox.OnSearchSelectShow()" + onmouseout="return searchBox.OnSearchSelectHide()" + onkeydown="return searchBox.OnSearchSelectKey(event)"> +</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> + +<div class="header"> + <div class="headertitle"> +<div class="title">Multinomial Logistic Regression<div class="ingroups"><a class="el" href="group__grp__deprecated.html">Deprecated Modules</a></div></div> </div> +</div><!--header--> +<div class="contents"> +<dl class="section warning"><dt>Warning</dt><dd><em> This is an old implementation of multinomial logistic regression. Replacement of this function is available as the Multinomial regression module <a class="el" href="group__grp__multinom.html">Multinomial Regression</a></em></dd></dl> +<div class="toc"><b>Contents</b> <ul> +<li class="level1"> +<a href="#train">Training Function</a> </li> +<li class="level1"> +<a href="#predict">Prediction Function</a> </li> +<li class="level1"> +<a href="#examples">Examples</a> </li> +<li class="level1"> +<a href="#background">Technical Background</a> </li> +<li class="level1"> +<a href="#literature">Literature</a> </li> +<li class="level1"> +<a href="#related">Related Topics</a> </li> +</ul> +</div><p>Multinomial logistic regression is a widely used regression analysis tool that models the outcomes of categorical dependent random variables. The model assumes that the conditional mean of the dependent categorical variables is the logistic function of an affine combination of independent variables. Multinomial logistic regression finds the vector of coefficients that maximizes the likelihood of the observations.</p> +<p><a class="anchor" id="train"></a></p><dl class="section user"><dt>Training Function</dt><dd>The multinomial logistic regression training function has the following syntax: <pre class="syntax"> +mlogregr_train(source_table, + output_table, + dependent_varname, + independent_varname, + ref_category, + optimizer_params + ) +</pre> <b>Arguments</b> <dl class="arglist"> +<dt>source_table </dt> +<dd><p class="startdd">TEXT. The name of the table containing the input data.</p> +<p></p> +<p class="enddd"></p> +</dd> +<dt>output_table </dt> +<dd><p class="startdd">TEXT. The name of the generated table containing the output model. The output table produced by the multinomial logistic regression training function contains the following columns: </p><table class="output"> +<tr> +<th>category </th><td>INTEGER. The category. Categories are encoded as integers with values from {0, 1, 2,..., <em>numCategories</em> – 1} </td></tr> +<tr> +<th>ref_category </th><td>INTEGER. The reference category. Categories are encoded as integers with values from {0, 1, 2,..., <em>numCategories</em> – 1} </td></tr> +<tr> +<th>coef </th><td>FLOAT8[]. An array of coefficients, \( \boldsymbol c \). </td></tr> +<tr> +<th>log_likelihood </th><td>FLOAT8. The log-likelihood, \( l(\boldsymbol c) \). </td></tr> +<tr> +<th>std_err </th><td>FLOAT8[]. An array of the standard errors. </td></tr> +<tr> +<th>z_stats </th><td>FLOAT8[]. An array of the Wald z-statistics. </td></tr> +<tr> +<th>p_values </th><td>FLOAT8[]. An array of the Wald p-values. </td></tr> +<tr> +<th>odds_ratios </th><td>FLOAT8[]. An array of the odds ratios. </td></tr> +<tr> +<th>condition_no </th><td>FLOAT8. The condition number of the matrix, computed using the coefficients of the iteration immediately preceding convergence. </td></tr> +<tr> +<th>num_iterations </th><td>INTEGER. The number of iterations executed before the algorithm completed. </td></tr> +</table> +<p>A summary table named <out_table>_summary is also created at the same time, and it contains the following columns:</p> +<table class="output"> +<tr> +<th>source_table </th><td>The data source table name. </td></tr> +<tr> +<th>out_table </th><td>The output table name. </td></tr> +<tr> +<th>dependent_varname </th><td>The dependent variable. </td></tr> +<tr> +<th>independent_varname </th><td>The independent variables. </td></tr> +<tr> +<th>optimizer_params </th><td>The optimizer parameters. It is a copy of the optimizer_params in the training function's arguments. </td></tr> +<tr> +<th>ref_category </th><td>An integer, the value of reference category used. </td></tr> +<tr> +<th>num_rows_processed </th><td>INTEGER. The number of rows actually processed, which is equal to the total number of rows in the source table minus the number of skipped rows. </td></tr> +<tr> +<th>num_missing_rows_skipped </th><td>INTEGER. The number of rows skipped during the training. A row will be skipped if the ind_col is NULL or contains NULL values. </td></tr> +</table> +<p class="enddd"></p> +</dd> +<dt>dependent_varname </dt> +<dd><p class="startdd">TEXT. The name of the column containing the dependent variable.</p> +<p class="enddd"></p> +</dd> +<dt>independent_varname </dt> +<dd><p class="startdd">TEXT. Expression list to evaluate for the independent variables. An intercept variable is not assumed. The number of independent variables cannot exceed 65535.</p> +<p class="enddd"></p> +</dd> +<dt>ref_category (optional) </dt> +<dd><p class="startdd">INTEGER, default: 0. The reference category ranges from [0, <em>numCategories</em> – 1].</p> +<p class="enddd"></p> +</dd> +<dt>optimizer_params (optional) </dt> +<dd>VARCHAR, default: NULL, which uses the default values of optimizer parameters. It should be a string that contains pairs of 'key=value' separated by commas. Supported parameters with their default values: max_iter=20, optimizer='irls', precision=1e-4. Currently, only 'irls' and 'newton' are allowed for 'optimizer'. </dd> +</dl> +</dd></dl> +<dl class="section note"><dt>Note</dt><dd>Table names can be optionally schema qualified and table and column names should follow the same case-sensitivity and quoting rules as in the database.</dd></dl> +<p><a class="anchor" id="predict"></a></p><dl class="section user"><dt>Prediction Function</dt><dd>The prediction function is provided to estimate the conditional mean given a new predictor. It has the following syntax: <pre class="syntax"> +mlogregr_predict( + model_table, + new_data_table, + id_col_name, + output_table, + type) +</pre></dd></dl> +<p><b>Arguments</b> </p><dl class="arglist"> +<dt>model_table </dt> +<dd><p class="startdd">TEXT. Name of the table containing the multilogistic model. This should be the output table returned from <em>mlogregr_train</em>.</p> +<p class="enddd"></p> +</dd> +<dt>new_data_table </dt> +<dd><p class="startdd">TEXT. Name of the table containing prediction data. This table is expected to contain the same features that were used during training. The table should also contain <em>id_col_name</em> used for identifying each row.</p> +<p class="enddd"></p> +</dd> +<dt>id_col_name </dt> +<dd><p class="startdd">TEXT. Name of the column containing id information in the source data. This is a mandatory argument and is used for correlating prediction table rows with the source. The values of this column are expected to be unique for each tuple. </p> +<p class="enddd"></p> +</dd> +<dt>output_table </dt> +<dd><p class="startdd">TEXT. Name of the table to output prediction results to. If this table already exists then an error is returned. This output table contains the <em>id_col_name</em> column giving the 'id' for each prediction.</p> +<p>If <em>type</em> = 'response', then the table has a single additional column with the prediction value of the response. The type of this column depends on the type of the response variable used during training.</p> +<p>If <em>type</em> = 'prob', then the table has multiple additional columns, one for each possible category. The columns are labeled as 'estimated_prob_<em>category_value</em>', where <em>category_value</em> represents the values of categories (0 to K-1).</p> +<p class="enddd"></p> +</dd> +<dt>type </dt> +<dd><p class="startdd">TEXT, optional, default: 'response'.</p> +<p>When <em>type</em> = 'prob', the probabilities of each category (including the reference category) is given.</p> +<p class="enddd">When <em>type</em> = 'response', a single output is provided which represents the prediction category for each tuple. This represents the category with the highest probability. </p> +</dd> +</dl> +<p><a class="anchor" id="examples"></a></p><dl class="section user"><dt>Examples</dt><dd></dd></dl> +<ol type="1"> +<li>Create the training data table. <pre class="example"> +DROP TABLE IF EXISTS test3; +CREATE TABLE test3 ( + feat1 INTEGER, + feat2 INTEGER, + cat INTEGER +); +INSERT INTO test3(feat1, feat2, cat) VALUES +(1,35,1), +(2,33,0), +(3,39,1), +(1,37,1), +(2,31,1), +(3,36,0), +(2,36,1), +(2,31,1), +(2,41,1), +(2,37,1), +(1,44,1), +(3,33,2), +(1,31,1), +(2,44,1), +(1,35,1), +(1,44,0), +(1,46,0), +(2,46,1), +(2,46,2), +(3,49,1), +(2,39,0), +(2,44,1), +(1,47,1), +(1,44,1), +(1,37,2), +(3,38,2), +(1,49,0), +(2,44,0), +(3,61,2), +(1,65,2), +(3,67,1), +(3,65,2), +(1,65,2), +(2,67,2), +(1,65,2), +(1,62,2), +(3,52,2), +(3,63,2), +(2,59,2), +(3,65,2), +(2,59,0), +(3,67,2), +(3,67,2), +(3,60,2), +(3,67,2), +(3,62,2), +(2,54,2), +(3,65,2), +(3,62,2), +(2,59,2), +(3,60,2), +(3,63,2), +(3,65,2), +(2,63,1), +(2,67,2), +(2,65,2), +(2,62,2); +</pre></li> +<li>Run the multilogistic regression function. <pre class="example"> +DROP TABLE IF EXISTS test3_output; +DROP TABLE IF EXISTS test3_output_summary; +SELECT madlib.mlogregr_train('test3', + 'test3_output', + 'cat', + 'ARRAY[1, feat1, feat2]', + 0, + 'max_iter=20, optimizer=irls, precision=0.0001' + ); +</pre></li> +<li>View the result: <pre class="example"> +-- Set extended display on for easier reading of output +\x on +SELECT * FROM test3_output; +</pre> Results: <pre class="result"> +-[ RECORD 1 ]--+------------------------------------------------------------ +category | 1 +ref_category | 0 +coef | {1.45474045211601,0.0849956182104023,-0.0172383499601956} +loglikelihood | -39.14759930999 +std_err | {2.13085072854143,0.585021661344715,0.0431487356292144} +z_stats | {0.682704063982831,0.145286275409074,-0.39950996729842} +p_values | {0.494793861210936,0.884484850387893,0.689517480964129} +odd_ratios | {4.28337158128448,1.08871229617973,0.982909380301134} +condition_no | 280069.034217586 +num_iterations | 5 +-[ RECORD 2 ]--+------------------------------------------------------------ +category | 2 +ref_category | 0 +coef | {-7.12908167688326,0.87648787696783,0.127886153027713} +loglikelihood | -39.14759930999 +std_err | {2.52104008297868,0.639575886323862,0.0445757462972303} +z_stats | {-2.82783352990566,1.37042045472615,2.86896269049475} +p_values | {0.00468641692252239,0.170555690550421,0.00411820373218956} +odd_ratios | {0.000801455044349486,2.40244718187161,1.13642361694409} +condition_no | 280069.034217586 +num_iterations | 5 +</pre></li> +<li>View all parameters used during the training <pre class="example"> +\x on +SELECT * FROM test3_output_summary; +</pre> Results: <pre class="result"> +-[ RECORD 1 ]------------+-------------------------------------------------- +method | mlogregr +source_table | test3 +out_table | test3_output +dependent_varname | cat +independent_varname | ARRAY[1, feat1, feat2] +optimizer_params | max_iter=20, optimizer=irls, precision=0.0001 +ref_category | 0 +num_categories | 3 +num_rows_processed | 57 +num_missing_rows_skipped | 0 +variance_covariance | {{4.54052482732554,3.01080140927409,-0.551901021610841,-0.380754019900586,-0.0784151362989211,-0.0510014701718268},{3.01080140927409,6.35564309998514,-0.351902272617974,-0.766730342510818,-0.051877550252329,-0.0954432017695571},{-0.551901021610841,-0.351902272617974,0.34225034424253,0.231740815080827,-0.00117521831508331,-0.00114043921343171},{-0.380754019900586,-0.766730342510818,0.231740815080827,0.409057314366954,-0.000556498286025567,-0.000404735750986327},{-0.0784151362989211,-0.051877550252329,-0.00117521831508331,-0.000556498286025569,0.00186181338639984,0.00121080293928445},{-0.0510014701718268,-0.0954432017695571,-0.00114043921343171,-0.000404735750986325,0.00121080293928446,0.00198699715795504}} +coef | {{1.45474045211601,0.0849956182104023,-0.0172383499601956},{-7.12908167688326,0.87648787696783,0.127886153027713}} +</pre></li> +</ol> +<p><a class="anchor" id="background"></a></p><dl class="section user"><dt>Technical Background</dt><dd>Multinomial logistic regression models the outcomes of categorical dependent random variables (denoted \( Y \in \{ 0,1,2 \ldots k \} \)). The model assumes that the conditional mean of the dependent categorical variables is the logistic function of an affine combination of independent variables (usually denoted \( \boldsymbol x \)). That is, <p class="formulaDsp"> +\[ E[Y \mid \boldsymbol x] = \sigma(\boldsymbol c^T \boldsymbol x) \] +</p> + for some unknown vector of coefficients \( \boldsymbol c \) and where \( \sigma(x) = \frac{1}{1 + \exp(-x)} \) is the logistic function. Multinomial logistic regression finds the vector of coefficients \( \boldsymbol c \) that maximizes the likelihood of the observations.</dd></dl> +<p>Let</p><ul> +<li>\( \boldsymbol y \in \{ 0,1 \}^{n \times k} \) denote the vector of observed dependent variables, with \( n \) rows and \( k \) columns, containing the observed values of the dependent variable,</li> +<li>\( X \in \mathbf R^{n \times k} \) denote the design matrix with \( k \) columns and \( n \) rows, containing all observed vectors of independent variables \( \boldsymbol x_i \) as rows.</li> +</ul> +<p>By definition, </p><p class="formulaDsp"> +\[ P[Y = y_i | \boldsymbol x_i] = \sigma((-1)^{y_i} \cdot \boldsymbol c^T \boldsymbol x_i) \,. \] +</p> +<p> Maximizing the likelihood \( \prod_{i=1}^n \Pr(Y = y_i \mid \boldsymbol x_i) \) is equivalent to maximizing the log-likelihood \( \sum_{i=1}^n \log \Pr(Y = y_i \mid \boldsymbol x_i) \), which simplifies to </p><p class="formulaDsp"> +\[ l(\boldsymbol c) = -\sum_{i=1}^n \log(1 + \exp((-1)^{y_i} \cdot \boldsymbol c^T \boldsymbol x_i)) \,. \] +</p> +<p> The Hessian of this objective is \( H = -X^T A X \) where \( A = \text{diag}(a_1, \dots, a_n) \) is the diagonal matrix with \( a_i = \sigma(\boldsymbol c^T \boldsymbol x) \cdot \sigma(-\boldsymbol c^T \boldsymbol x) \,. \) Since \( H \) is non-positive definite, \( l(\boldsymbol c) \) is convex. There are many techniques for solving convex optimization problems. Currently, logistic regression in MADlib can use:</p><ul> +<li>Iteratively Reweighted Least Squares</li> +</ul> +<p>We estimate the standard error for coefficient \( i \) as </p><p class="formulaDsp"> +\[ \mathit{se}(c_i) = \left( (X^T A X)^{-1} \right)_{ii} \,. \] +</p> +<p> The Wald z-statistic is </p><p class="formulaDsp"> +\[ z_i = \frac{c_i}{\mathit{se}(c_i)} \,. \] +</p> +<p>The Wald \( p \)-value for coefficient \( i \) gives the probability (under the assumptions inherent in the Wald test) of seeing a value at least as extreme as the one observed, provided that the null hypothesis ( \( c_i = 0 \)) is true. Letting \( F \) denote the cumulative density function of a standard normal distribution, the Wald \( p \)-value for coefficient \( i \) is therefore </p><p class="formulaDsp"> +\[ p_i = \Pr(|Z| \geq |z_i|) = 2 \cdot (1 - F( |z_i| )) \] +</p> +<p> where \( Z \) is a standard normally distributed random variable.</p> +<p>The odds ratio for coefficient \( i \) is estimated as \( \exp(c_i) \).</p> +<p>The condition number is computed as \( \kappa(X^T A X) \) during the iteration immediately <em>preceding</em> convergence (i.e., \( A \) is computed using the coefficients of the previous iteration). A large condition number (say, more than 1000) indicates the presence of significant multicollinearity.</p> +<p>The multinomial logistic regression uses a default reference category of zero, and the regression coefficients in the output are in the order described below. For a problem with \( K \) dependent variables \( (1, ..., K) \) and \( J \) categories \( (0, ..., J-1) \), let \( {m_{k,j}} \) denote the coefficient for dependent variable \( k \) and category \( j \). The output is \( {m_{k_1, j_0}, m_{k_1, j_1} \ldots m_{k_1, j_{J-1}}, m_{k_2, j_0}, m_{k_2, j_1}, \ldots m_{k_2, j_{J-1}} \ldots m_{k_K, j_{J-1}}} \). The order is NOT CONSISTENT with the multinomial regression marginal effect calculation with function <em>marginal_mlogregr</em>. This is deliberate because the interfaces of all multinomial regressions (robust, clustered, ...) will be moved to match that used in marginal.</p> +<p><a class="anchor" id="literature"></a></p><dl class="section user"><dt>Literature</dt><dd></dd></dl> +<p>A collection of nice write-ups, with valuable pointers into further literature:</p> +<p>[1] Annette J. Dobson: An Introduction to Generalized Linear Models, Second Edition. Nov 2001</p> +<p>[2] Cosma Shalizi: Statistics 36-350: Data Mining, Lecture Notes, 18 November 2009, <a href="http://www.stat.cmu.edu/~cshalizi/350/lectures/26/lecture-26.pdf";>http://www.stat.cmu.edu/~cshalizi/350/lectures/26/lecture-26.pdf</a></p> +<p>[3] Scott A. Czepiel: Maximum Likelihood Estimation of Logistic Regression Models: Theory and Implementation, Retrieved Jul 12 2012, <a href="http://czep.net/stat/mlelr.pdf";>http://czep.net/stat/mlelr.pdf</a></p> +<p><a class="anchor" id="related"></a></p><dl class="section user"><dt>Related Topics</dt><dd></dd></dl> +<p>File <a class="el" href="multilogistic_8sql__in.html" title="SQL functions for multinomial logistic regression. ">multilogistic.sql_in</a> documenting the multinomial logistic regression functions</p> +<p><a class="el" href="group__grp__logreg.html">Logistic Regression</a></p> +</div><!-- contents --> +</div><!-- doc-content --> +<!-- start footer part --> +<div id="nav-path" class="navpath"><!-- id is needed for treeview function! --> + <ul> + <li class="footer">Generated on Mon Oct 15 2018 11:24:30 for MADlib by + <a href="http://www.doxygen.org/index.html";> + <img class="footer" src="doxygen.png" alt="doxygen"/></a> 1.8.14 </li> + </ul> +</div> +</body> +</html> http://git-wip-us.apache.org/repos/asf/madlib-site/blob/af0e5f14/docs/v1.15.1/group__grp__multinom.html ---------------------------------------------------------------------- diff --git a/docs/v1.15.1/group__grp__multinom.html b/docs/v1.15.1/group__grp__multinom.html new file mode 100644 index 0000000..9081d96 --- /dev/null +++ b/docs/v1.15.1/group__grp__multinom.html @@ -0,0 +1,494 @@ +<!-- HTML header for doxygen 1.8.4--> +<!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"/> +<meta http-equiv="X-UA-Compatible" content="IE=9"/> +<meta name="generator" content="Doxygen 1.8.14"/> +<meta name="keywords" content="madlib,postgres,greenplum,machine learning,data mining,deep learning,ensemble methods,data science,market basket analysis,affinity analysis,pca,lda,regression,elastic net,huber white,proportional hazards,k-means,latent dirichlet allocation,bayes,support vector machines,svm"/> +<title>MADlib: Multinomial Regression</title> +<link href="tabs.css" rel="stylesheet" type="text/css"/> +<script type="text/javascript" src="jquery.js"></script> +<script type="text/javascript" src="dynsections.js"></script> +<link href="navtree.css" rel="stylesheet" type="text/css"/> +<script type="text/javascript" src="resize.js"></script> +<script type="text/javascript" src="navtreedata.js"></script> +<script type="text/javascript" src="navtree.js"></script> +<script type="text/javascript"> +/* @license magnet:?xt=urn:btih:cf05388f2679ee054f2beb29a391d25f4e673ac3&dn=gpl-2.0.txt GPL-v2 */ + $(document).ready(initResizable); +/* @license-end */</script> +<link href="search/search.css" rel="stylesheet" type="text/css"/> +<script type="text/javascript" src="search/searchdata.js"></script> +<script type="text/javascript" src="search/search.js"></script> +<script type="text/javascript"> +/* @license magnet:?xt=urn:btih:cf05388f2679ee054f2beb29a391d25f4e673ac3&dn=gpl-2.0.txt GPL-v2 */ + $(document).ready(function() { init_search(); }); +/* @license-end */ +</script> +<script type="text/x-mathjax-config"> + MathJax.Hub.Config({ + extensions: ["tex2jax.js", "TeX/AMSmath.js", "TeX/AMSsymbols.js"], + jax: ["input/TeX","output/HTML-CSS"], +}); +</script><script type="text/javascript" async src="https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/MathJax.js";></script> +<!-- hack in the navigation tree --> +<script type="text/javascript" src="eigen_navtree_hacks.js"></script> +<link href="doxygen.css" rel="stylesheet" type="text/css" /> +<link href="madlib_extra.css" rel="stylesheet" type="text/css"/> +<!-- google analytics --> +<script> + (function(i,s,o,g,r,a,m){i['GoogleAnalyticsObject']=r;i[r]=i[r]||function(){ + (i[r].q=i[r].q||[]).push(arguments)},i[r].l=1*new Date();a=s.createElement(o), + m=s.getElementsByTagName(o)[0];a.async=1;a.src=g;m.parentNode.insertBefore(a,m) + })(window,document,'script','//www.google-analytics.com/analytics.js','ga'); + ga('create', 'UA-45382226-1', 'madlib.apache.org'); + ga('send', 'pageview'); +</script> +</head> +<body> +<div id="top"><!-- do not remove this div, it is closed by doxygen! --> +<div id="titlearea"> +<table cellspacing="0" cellpadding="0"> + <tbody> + <tr style="height: 56px;"> + <td id="projectlogo"><a href="http://madlib.apache.org";><img alt="Logo" src="madlib.png" height="50" style="padding-left:0.5em;" border="0"/ ></a></td> + <td style="padding-left: 0.5em;"> + <div id="projectname"> + <span id="projectnumber">1.15.1</span> + </div> + <div id="projectbrief">User Documentation for Apache MADlib</div> + </td> + <td> <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> +</td> + </tr> + </tbody> +</table> +</div> +<!-- end header part --> +<!-- Generated by Doxygen 1.8.14 --> +<script type="text/javascript"> +/* @license magnet:?xt=urn:btih:cf05388f2679ee054f2beb29a391d25f4e673ac3&dn=gpl-2.0.txt GPL-v2 */ +var searchBox = new SearchBox("searchBox", "search",false,'Search'); +/* @license-end */ +</script> +</div><!-- top --> +<div id="side-nav" class="ui-resizable side-nav-resizable"> + <div id="nav-tree"> + <div id="nav-tree-contents"> + <div id="nav-sync" class="sync"></div> + </div> + </div> + <div id="splitbar" style="-moz-user-select:none;" + class="ui-resizable-handle"> + </div> +</div> +<script type="text/javascript"> +/* @license magnet:?xt=urn:btih:cf05388f2679ee054f2beb29a391d25f4e673ac3&dn=gpl-2.0.txt GPL-v2 */ +$(document).ready(function(){initNavTree('group__grp__multinom.html','');}); +/* @license-end */ +</script> +<div id="doc-content"> +<!-- window showing the filter options --> +<div id="MSearchSelectWindow" + onmouseover="return searchBox.OnSearchSelectShow()" + onmouseout="return searchBox.OnSearchSelectHide()" + onkeydown="return searchBox.OnSearchSelectKey(event)"> +</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> + +<div class="header"> + <div class="headertitle"> +<div class="title">Multinomial Regression<div class="ingroups"><a class="el" href="group__grp__super.html">Supervised Learning</a> » <a class="el" href="group__grp__regml.html">Regression Models</a></div></div> </div> +</div><!--header--> +<div class="contents"> +<div class="toc"><b>Contents</b> <ul> +<li class="level1"> +<a href="#train">Training Function</a> </li> +<li class="level1"> +<a href="#predict">Prediction Function</a> </li> +<li class="level1"> +<a href="#examples">Examples</a> </li> +<li class="level1"> +<a href="#background">Technical Background</a> </li> +<li class="level1"> +<a href="#literature">Literature</a> </li> +<li class="level1"> +<a href="#related">Related Topics</a> </li> +</ul> +</div><p>In statistics, multinomial regression is a classification method that generalizes binomial regression to multiclass problems, i.e. with more than two possible discrete outcomes. That is, it is a model that is used to predict the probabilities of the different possible outcomes of a categorically distributed dependent variable, given a set of independent variables (which may be real-valued, binary-valued, categorical-valued, etc.).</p> +<p><a class="anchor" id="train"></a></p><dl class="section user"><dt>Training Function</dt><dd>The multinomial regression training function has the following syntax: <pre class="syntax"> +multinom(source_table, + model_table, + dependent_varname, + independent_varname, + ref_category, + link_func, + grouping_col, + optim_params, + verbose + ) +</pre></dd></dl> +<p><b>Arguments</b> </p><dl class="arglist"> +<dt>source_table </dt> +<dd><p class="startdd">VARCHAR. Name of the table containing the training data.</p> +<p class="enddd"></p> +</dd> +<dt>model_table </dt> +<dd><p class="startdd">VARCHAR. Name of the generated table containing the model.</p> +<p>The model table produced by multinom() contains the following columns:</p> +<table class="output"> +<tr> +<th><...> </th><td><p class="starttd">Grouping columns, if provided in input. This could be multiple columns depending on the <code>grouping_col</code> input. </p> +<p class="endtd"></p> +</td></tr> +<tr> +<th>category </th><td><p class="starttd">VARCHAR. String representation of category value. </p> +<p class="endtd"></p> +</td></tr> +<tr> +<th>coef </th><td><p class="starttd">FLOAT8[]. Vector of the coefficients in linear predictor. </p> +<p class="endtd"></p> +</td></tr> +<tr> +<th>log_likelihood </th><td><p class="starttd">FLOAT8. The log-likelihood \( l(\boldsymbol \beta) \). The value will be the same across categories within the same group. </p> +<p class="endtd"></p> +</td></tr> +<tr> +<th>std_err </th><td><p class="starttd">FLOAT8[]. Vector of the standard errors of the coefficients. </p> +<p class="endtd"></p> +</td></tr> +<tr> +<th>z_stats </th><td><p class="starttd">FLOAT8[]. Vector of the z-statistics of the coefficients. </p> +<p class="endtd"></p> +</td></tr> +<tr> +<th>p_values </th><td><p class="starttd">FLOAT8[]. Vector of the p-values of the coefficients. </p> +<p class="endtd"></p> +</td></tr> +<tr> +<th>num_rows_processed </th><td><p class="starttd">BIGINT. Number of rows processed. </p> +<p class="endtd"></p> +</td></tr> +<tr> +<th>num_rows_skipped </th><td><p class="starttd">BIGINT. Number of rows skipped due to missing values or failures. </p> +<p class="endtd"></p> +</td></tr> +<tr> +<th>num_iterations </th><td>INTEGER. Number of iterations actually completed. This would be different from the <code>nIterations</code> argument if a <code>tolerance</code> parameter is provided and the algorithm converges before all iterations are completed. </td></tr> +</table> +<p>A summary table named <model_table>_summary is also created at the same time, which has the following columns: </p><table class="output"> +<tr> +<th>method </th><td><p class="starttd">VARCHAR. String describes the model: 'multinom'. </p> +<p class="endtd"></p> +</td></tr> +<tr> +<th>source_table </th><td><p class="starttd">VARCHAR. Data source table name. </p> +<p class="endtd"></p> +</td></tr> +<tr> +<th>model_table </th><td><p class="starttd">VARCHAR. Model table name. </p> +<p class="endtd"></p> +</td></tr> +<tr> +<th>dependent_varname </th><td><p class="starttd">VARCHAR. Expression for dependent variable. </p> +<p class="endtd"></p> +</td></tr> +<tr> +<th>independent_varname </th><td><p class="starttd">VARCHAR. Expression for independent variables. </p> +<p class="endtd"></p> +</td></tr> +<tr> +<th>ref_category </th><td><p class="starttd">VARCHAR. String representation of reference category. </p> +<p class="endtd"></p> +</td></tr> +<tr> +<th>link_func </th><td><p class="starttd">VARCHAR. String that contains link function parameters: only 'logit' is implemented now </p> +<p class="endtd"></p> +</td></tr> +<tr> +<th>grouping_col </th><td><p class="starttd">VARCHAR. String representation of grouping columns. </p> +<p class="endtd"></p> +</td></tr> +<tr> +<th>optimizer_params </th><td><p class="starttd">VARCHAR. String that contains optimizer parameters, and has the form of 'optimizer=..., max_iter=..., tolerance=...'. </p> +<p class="endtd"></p> +</td></tr> +<tr> +<th>num_all_groups </th><td><p class="starttd">INTEGER. Number of groups in glm training. </p> +<p class="endtd"></p> +</td></tr> +<tr> +<th>num_failed_groups </th><td><p class="starttd">INTEGER. Number of failed groups in glm training. </p> +<p class="endtd"></p> +</td></tr> +<tr> +<th>total_rows_processed </th><td><p class="starttd">BIGINT. Total number of rows processed in all groups. </p> +<p class="endtd"></p> +</td></tr> +<tr> +<th>total_rows_skipped </th><td><p class="starttd">BIGINT. Total number of rows skipped in all groups due to missing values or failures. </p> +<p class="endtd"></p> +</td></tr> +</table> +<p class="enddd"></p> +</dd> +<dt>dependent_varname </dt> +<dd><p class="startdd">VARCHAR. Name of the dependent variable column.</p> +<p class="enddd"></p> +</dd> +<dt>independent_varname </dt> +<dd><p class="startdd">VARCHAR. Expression list to evaluate for the independent variables. An intercept variable is not assumed. It is common to provide an explicit intercept term by including a single constant <code>1</code> term in the independent variable list.</p> +<p class="enddd"></p> +</dd> +<dt>link_function (optional) </dt> +<dd><p class="startdd">VARCHAR, default: 'logit'. Parameters for link function. Currently, we support logit. </p> +<p class="enddd"></p> +</dd> +<dt>ref_category (optional) </dt> +<dd><p class="startdd">VARCHAR, default: '0'. Parameters to specify the reference category. </p> +<p class="enddd"></p> +</dd> +<dt>grouping_col (optional) </dt> +<dd><p class="startdd">VARCHAR, default: NULL. An expression list used to group the input dataset into discrete groups, running one regression per group. Similar to the SQL "GROUP BY" clause. When this value is NULL, no grouping is used and a single model is generated.</p> +<p class="enddd"></p> +</dd> +<dt>optim_params (optional) </dt> +<dd><p class="startdd">VARCHAR, default: 'max_iter=100,optimizer=irls,tolerance=1e-6'. Parameters for optimizer. Currently, we support tolerance=[tolerance for relative error between log-likelihoods], max_iter=[maximum iterations to run], optimizer=irls.</p> +<p class="enddd"></p> +</dd> +<dt>verbose (optional) </dt> +<dd>BOOLEAN, default: FALSE. Provides verbose output of the results of training. </dd> +</dl> +<dl class="section note"><dt>Note</dt><dd>For p-values, we just return the computation result directly. Other statistical packages, like 'R', produce the same result, but on printing the result to screen, another format function is used and any p-value that is smaller than the machine epsilon (the smallest positive floating-point number 'x' such that '1 + x != 1') will be printed on screen as "< xxx" (xxx is the value of the machine epsilon). Although the result may look different, they are in fact the same. </dd></dl> +<p><a class="anchor" id="predict"></a></p><dl class="section user"><dt>Prediction Function</dt><dd>Multinomial regression prediction function has the following format: <pre class="syntax"> +multinom_predict(model_table, + predict_table_input, + output_table, + predict_type, + verbose, + id_column + ) +</pre> <b>Arguments</b> <dl class="arglist"> +<dt>model_table </dt> +<dd><p class="startdd">TEXT. Name of the generated table containing the model, which is the output table from multinom().</p> +<p class="enddd"></p> +</dd> +<dt>predict_table_input </dt> +<dd><p class="startdd">TEXT. The name of the table containing the data to predict on. The table must contain id column as the primary key.</p> +<p class="enddd"></p> +</dd> +<dt>output_table </dt> +<dd><p class="startdd">TEXT. Name of the generated table containing the predicted values.</p> +<p>The model table produced by multinom_predict contains the following columns:</p> +<table class="output"> +<tr> +<th>id </th><td><p class="starttd">SERIAL. Column to identify the predicted value. </p> +<p class="endtd"></p> +</td></tr> +<tr> +<th>category </th><td><p class="starttd">TEXT. Available if the predicted type = 'response'. Column contains the predicted categories </p> +<p class="endtd"></p> +</td></tr> +<tr> +<th>category_value </th><td>FLOAT8. The predicted probability for the specific category_value. </td></tr> +</table> +<p class="enddd"></p> +</dd> +<dt>predict_type </dt> +<dd>TEXT. Either 'response' or 'probability'. Using 'response' will give the predicted category with the largest probability. Using probability will give the predicted probabilities for all categories </dd> +<dt>verbose </dt> +<dd><p class="startdd">BOOLEAN. Control whether verbose is displayed. The default is FALSE. </p> +<p class="enddd"></p> +</dd> +<dt>id_column </dt> +<dd>TEXT. The name of the column in the input table. </dd> +</dl> +</dd></dl> +<p><a class="anchor" id="examples"></a></p><dl class="section user"><dt>Examples</dt><dd></dd></dl> +<ol type="1"> +<li>Create the training data table. <pre class="example"> +DROP TABLE IF EXISTS test3; +CREATE TABLE test3 ( + feat1 INTEGER, + feat2 INTEGER, + cat INTEGER +); +INSERT INTO test3(feat1, feat2, cat) VALUES +(1,35,1), +(2,33,0), +(3,39,1), +(1,37,1), +(2,31,1), +(3,36,0), +(2,36,1), +(2,31,1), +(2,41,1), +(2,37,1), +(1,44,1), +(3,33,2), +(1,31,1), +(2,44,1), +(1,35,1), +(1,44,0), +(1,46,0), +(2,46,1), +(2,46,2), +(3,49,1), +(2,39,0), +(2,44,1), +(1,47,1), +(1,44,1), +(1,37,2), +(3,38,2), +(1,49,0), +(2,44,0), +(3,61,2), +(1,65,2), +(3,67,1), +(3,65,2), +(1,65,2), +(2,67,2), +(1,65,2), +(1,62,2), +(3,52,2), +(3,63,2), +(2,59,2), +(3,65,2), +(2,59,0), +(3,67,2), +(3,67,2), +(3,60,2), +(3,67,2), +(3,62,2), +(2,54,2), +(3,65,2), +(3,62,2), +(2,59,2), +(3,60,2), +(3,63,2), +(3,65,2), +(2,63,1), +(2,67,2), +(2,65,2), +(2,62,2); +</pre></li> +<li>Run the multilogistic regression function. <pre class="example"> +DROP TABLE IF EXISTS test3_output; +DROP TABLE IF EXISTS test3_output_summary; +SELECT madlib.multinom('test3', + 'test3_output', + 'cat', + 'ARRAY[1, feat1, feat2]', + '0', + 'logit' + ); +</pre></li> +<li>View the regression results. <pre class="example"> +-- Set extended display on for easier reading of output +\x on +SELECT * FROM test3_output; +</pre></li> +</ol> +<p>Result: </p><pre class="result"> +-[ RECORD 1 ]------+------------------------------------------------------------ +category | 1 +coef | {1.45474045165731,0.084995618282504,-0.0172383499512136} +log_likelihood | -39.1475993094045 +std_err | {2.13085878785549,0.585023211942952,0.0431489262260687} +z_stats | {0.682701481650677,0.145285890452484,-0.399508202380224} +p_values | {0.494795493298706,0.884485154314181,0.689518781152604} +num_rows_processed | 57 +num_rows_skipped | 0 +iteration | 6 +-[ RECORD 2 ]------+------------------------------------------------------------ +category | 2 +coef | {-7.1290816775109,0.876487877074751,0.127886153038661} +log_likelihood | -39.1475993094045 +std_err | {2.52105418324135,0.639578886139654,0.0445760103748678} +z_stats | {-2.82781771407425,1.37041402721253,2.86894569440347} +p_values | {0.00468664844488755,0.170557695812408,0.00411842502754068} +num_rows_processed | 57 +num_rows_skipped | 0 +iteration | 6 +</pre><ol type="1"> +<li>Predicting dependent variable using multinomial model. (This example uses the original data table to perform the prediction. Typically a different test dataset with the same features as the original training dataset would be used for prediction.)</li> +</ol> +<pre class="example"> +\x off +-- Add the id column for prediction function +ALTER TABLE test3 ADD COLUMN id SERIAL; +-- Predict probabilities for all categories using the original data +SELECT madlib.multinom_predict('test3_out','test3', 'test3_prd_prob', 'probability'); +-- Display the predicted value +SELECT * FROM test3_prd_prob; +</pre><p><a class="anchor" id="background"></a></p><dl class="section user"><dt>Technical Background</dt><dd>When link = 'logit', multinomial logistic regression models the outcomes of categorical dependent random variables (denoted \( Y \in \{ 0,1,2 \ldots k \} \)). The model assumes that the conditional mean of the dependent categorical variables is the logistic function of an affine combination of independent variables (usually denoted \( \boldsymbol x \)). That is, <p class="formulaDsp"> +\[ E[Y \mid \boldsymbol x] = \sigma(\boldsymbol c^T \boldsymbol x) \] +</p> + for some unknown vector of coefficients \( \boldsymbol c \) and where \( \sigma(x) = \frac{1}{1 + \exp(-x)} \) is the logistic function. Multinomial logistic regression finds the vector of coefficients \( \boldsymbol c \) that maximizes the likelihood of the observations.</dd></dl> +<p>Let</p><ul> +<li>\( \boldsymbol y \in \{ 0,1 \}^{n \times k} \) denote the vector of observed dependent variables, with \( n \) rows and \( k \) columns, containing the observed values of the dependent variable,</li> +<li>\( X \in \mathbf R^{n \times k} \) denote the design matrix with \( k \) columns and \( n \) rows, containing all observed vectors of independent variables \( \boldsymbol x_i \) as rows.</li> +</ul> +<p>By definition, </p><p class="formulaDsp"> +\[ P[Y = y_i | \boldsymbol x_i] = \sigma((-1)^{y_i} \cdot \boldsymbol c^T \boldsymbol x_i) \,. \] +</p> +<p> Maximizing the likelihood \( \prod_{i=1}^n \Pr(Y = y_i \mid \boldsymbol x_i) \) is equivalent to maximizing the log-likelihood \( \sum_{i=1}^n \log \Pr(Y = y_i \mid \boldsymbol x_i) \), which simplifies to </p><p class="formulaDsp"> +\[ l(\boldsymbol c) = -\sum_{i=1}^n \log(1 + \exp((-1)^{y_i} \cdot \boldsymbol c^T \boldsymbol x_i)) \,. \] +</p> +<p> The Hessian of this objective is \( H = -X^T A X \) where \( A = \text{diag}(a_1, \dots, a_n) \) is the diagonal matrix with \( a_i = \sigma(\boldsymbol c^T \boldsymbol x) \cdot \sigma(-\boldsymbol c^T \boldsymbol x) \,. \) Since \( H \) is non-positive definite, \( l(\boldsymbol c) \) is convex. There are many techniques for solving convex optimization problems. Currently, logistic regression in MADlib can use:</p><ul> +<li>Iteratively Reweighted Least Squares</li> +</ul> +<p>We estimate the standard error for coefficient \( i \) as </p><p class="formulaDsp"> +\[ \mathit{se}(c_i) = \left( (X^T A X)^{-1} \right)_{ii} \,. \] +</p> +<p> The Wald z-statistic is </p><p class="formulaDsp"> +\[ z_i = \frac{c_i}{\mathit{se}(c_i)} \,. \] +</p> +<p>The Wald \( p \)-value for coefficient \( i \) gives the probability (under the assumptions inherent in the Wald test) of seeing a value at least as extreme as the one observed, provided that the null hypothesis ( \( c_i = 0 \)) is true. Letting \( F \) denote the cumulative density function of a standard normal distribution, the Wald \( p \)-value for coefficient \( i \) is therefore </p><p class="formulaDsp"> +\[ p_i = \Pr(|Z| \geq |z_i|) = 2 \cdot (1 - F( |z_i| )) \] +</p> +<p> where \( Z \) is a standard normally distributed random variable.</p> +<p>The odds ratio for coefficient \( i \) is estimated as \( \exp(c_i) \).</p> +<p>The condition number is computed as \( \kappa(X^T A X) \) during the iteration immediately <em>preceding</em> convergence (i.e., \( A \) is computed using the coefficients of the previous iteration). A large condition number (say, more than 1000) indicates the presence of significant multicollinearity.</p> +<p>The multinomial logistic regression uses a default reference category of zero, and the regression coefficients in the output are in the order described below. For a problem with \( K \) dependent variables \( (1, ..., K) \) and \( J \) categories \( (0, ..., J-1) \), let \( {m_{k,j}} \) denote the coefficient for dependent variable \( k \) and category \( j \). The output is \( {m_{k_1, j_0}, m_{k_1, j_1} \ldots m_{k_1, j_{J-1}}, m_{k_2, j_0}, m_{k_2, j_1}, \ldots m_{k_2, j_{J-1}} \ldots m_{k_K, j_{J-1}}} \). The order is NOT CONSISTENT with the multinomial regression marginal effect calculation with function <em>marginal_mlogregr</em>. This is deliberate because the interfaces of all multinomial regressions (robust, clustered, ...) will be moved to match that used in marginal.</p> +<p><a class="anchor" id="literature"></a></p><dl class="section user"><dt>Literature</dt><dd></dd></dl> +<p>A collection of nice write-ups, with valuable pointers into further literature:</p> +<p>[1] Annette J. Dobson: An Introduction to Generalized Linear Models, Second Edition. Nov 2001</p> +<p>[2] Cosma Shalizi: Statistics 36-350: Data Mining, Lecture Notes, 18 November 2009, <a href="http://www.stat.cmu.edu/~cshalizi/350/lectures/26/lecture-26.pdf";>http://www.stat.cmu.edu/~cshalizi/350/lectures/26/lecture-26.pdf</a></p> +<p>[3] Scott A. Czepiel: Maximum Likelihood Estimation of Logistic Regression Models: Theory and Implementation, Retrieved Jul 12 2012, <a href="http://czep.net/stat/mlelr.pdf";>http://czep.net/stat/mlelr.pdf</a></p> +<p><a class="anchor" id="related"></a></p><dl class="section user"><dt>Related Topics</dt><dd></dd></dl> +<p>File <a class="el" href="multiresponseglm_8sql__in.html" title="SQL functions for multinomial regression. ">multiresponseglm.sql_in</a> documenting the multinomial regression functions</p> +<p><a class="el" href="group__grp__logreg.html">Logistic Regression</a></p> +<p><a class="el" href="group__grp__ordinal.html">Ordinal Regression</a></p> +</div><!-- contents --> +</div><!-- doc-content --> +<!-- start footer part --> +<div id="nav-path" class="navpath"><!-- id is needed for treeview function! --> + <ul> + <li class="footer">Generated on Mon Oct 15 2018 11:24:30 for MADlib by + <a href="http://www.doxygen.org/index.html";> + <img class="footer" src="doxygen.png" alt="doxygen"/></a> 1.8.14 </li> + </ul> +</div> +</body> +</html>
