361 lines
45 KiB
HTML
361 lines
45 KiB
HTML
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "https://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
|
||
<html xmlns="http://www.w3.org/1999/xhtml" lang="en-US">
|
||
<head>
|
||
<meta http-equiv="Content-Type" content="text/xhtml;charset=UTF-8"/>
|
||
<meta http-equiv="X-UA-Compatible" content="IE=11"/>
|
||
<meta name="generator" content="Doxygen 1.11.0"/>
|
||
<meta name="viewport" content="width=device-width, initial-scale=1"/>
|
||
<title>BayesNet: /Users/rmontanana/Code/BayesNet/bayesnet/ensembles/Boost.cc Source File</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>
|
||
<script type="text/javascript" src="clipboard.js"></script>
|
||
<link href="navtree.css" rel="stylesheet" type="text/css"/>
|
||
<script type="text/javascript" src="navtreedata.js"></script>
|
||
<script type="text/javascript" src="navtree.js"></script>
|
||
<script type="text/javascript" src="resize.js"></script>
|
||
<script type="text/javascript" src="cookie.js"></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>
|
||
<link href="doxygen.css" rel="stylesheet" type="text/css" />
|
||
</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 id="projectrow">
|
||
<td id="projectlogo"><img alt="Logo" src="logo_small.png"/></td>
|
||
<td id="projectalign">
|
||
<div id="projectname">BayesNet<span id="projectnumber"> 1.0.5</span>
|
||
</div>
|
||
<div id="projectbrief">Bayesian Network Classifiers using libtorch from scratch</div>
|
||
</td>
|
||
</tr>
|
||
</tbody>
|
||
</table>
|
||
</div>
|
||
<!-- end header part -->
|
||
<!-- Generated by Doxygen 1.11.0 -->
|
||
<script type="text/javascript">
|
||
/* @license magnet:?xt=urn:btih:d3d9a9a6595521f9666a5e94cc830dab83b65699&dn=expat.txt MIT */
|
||
var searchBox = new SearchBox("searchBox", "search/",'.html');
|
||
/* @license-end */
|
||
</script>
|
||
<script type="text/javascript">
|
||
/* @license magnet:?xt=urn:btih:d3d9a9a6595521f9666a5e94cc830dab83b65699&dn=expat.txt MIT */
|
||
$(function() { codefold.init(0); });
|
||
/* @license-end */
|
||
</script>
|
||
<script type="text/javascript" src="menudata.js"></script>
|
||
<script type="text/javascript" src="menu.js"></script>
|
||
<script type="text/javascript">
|
||
/* @license magnet:?xt=urn:btih:d3d9a9a6595521f9666a5e94cc830dab83b65699&dn=expat.txt MIT */
|
||
$(function() {
|
||
initMenu('',true,false,'search.php','Search',true);
|
||
$(function() { init_search(); });
|
||
});
|
||
/* @license-end */
|
||
</script>
|
||
<div id="main-nav"></div>
|
||
</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:d3d9a9a6595521f9666a5e94cc830dab83b65699&dn=expat.txt MIT */
|
||
$(function(){initNavTree('_boost_8cc_source.html',''); initResizable(true); });
|
||
/* @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">
|
||
<div id="MSearchResults">
|
||
<div class="SRPage">
|
||
<div id="SRIndex">
|
||
<div id="SRResults"></div>
|
||
<div class="SRStatus" id="Loading">Loading...</div>
|
||
<div class="SRStatus" id="Searching">Searching...</div>
|
||
<div class="SRStatus" id="NoMatches">No Matches</div>
|
||
</div>
|
||
</div>
|
||
</div>
|
||
</div>
|
||
|
||
<div class="header">
|
||
<div class="headertitle"><div class="title">Boost.cc</div></div>
|
||
</div><!--header-->
|
||
<div class="contents">
|
||
<div class="fragment"><div class="line"><a id="l00001" name="l00001"></a><span class="lineno"> 1</span><span class="comment">// ***************************************************************</span></div>
|
||
<div class="line"><a id="l00002" name="l00002"></a><span class="lineno"> 2</span><span class="comment">// SPDX-FileCopyrightText: Copyright 2024 Ricardo Montañana Gómez</span></div>
|
||
<div class="line"><a id="l00003" name="l00003"></a><span class="lineno"> 3</span><span class="comment">// SPDX-FileType: SOURCE</span></div>
|
||
<div class="line"><a id="l00004" name="l00004"></a><span class="lineno"> 4</span><span class="comment">// SPDX-License-Identifier: MIT</span></div>
|
||
<div class="line"><a id="l00005" name="l00005"></a><span class="lineno"> 5</span><span class="comment">// ***************************************************************</span></div>
|
||
<div class="line"><a id="l00006" name="l00006"></a><span class="lineno"> 6</span><span class="preprocessor">#include <folding.hpp></span></div>
|
||
<div class="line"><a id="l00007" name="l00007"></a><span class="lineno"> 7</span><span class="preprocessor">#include "bayesnet/feature_selection/CFS.h"</span></div>
|
||
<div class="line"><a id="l00008" name="l00008"></a><span class="lineno"> 8</span><span class="preprocessor">#include "bayesnet/feature_selection/FCBF.h"</span></div>
|
||
<div class="line"><a id="l00009" name="l00009"></a><span class="lineno"> 9</span><span class="preprocessor">#include "bayesnet/feature_selection/IWSS.h"</span></div>
|
||
<div class="line"><a id="l00010" name="l00010"></a><span class="lineno"> 10</span><span class="preprocessor">#include "Boost.h"</span></div>
|
||
<div class="line"><a id="l00011" name="l00011"></a><span class="lineno"> 11</span> </div>
|
||
<div class="line"><a id="l00012" name="l00012"></a><span class="lineno"> 12</span><span class="keyword">namespace </span>bayesnet {</div>
|
||
<div class="line"><a id="l00013" name="l00013"></a><span class="lineno"> 13</span> Boost::Boost(<span class="keywordtype">bool</span> predict_voting) : Ensemble(predict_voting)</div>
|
||
<div class="line"><a id="l00014" name="l00014"></a><span class="lineno"> 14</span> {</div>
|
||
<div class="line"><a id="l00015" name="l00015"></a><span class="lineno"> 15</span> validHyperparameters = { <span class="stringliteral">"order"</span>, <span class="stringliteral">"convergence"</span>, <span class="stringliteral">"convergence_best"</span>, <span class="stringliteral">"bisection"</span>, <span class="stringliteral">"threshold"</span>, <span class="stringliteral">"maxTolerance"</span>,</div>
|
||
<div class="line"><a id="l00016" name="l00016"></a><span class="lineno"> 16</span> <span class="stringliteral">"predict_voting"</span>, <span class="stringliteral">"select_features"</span>, <span class="stringliteral">"block_update"</span> };</div>
|
||
<div class="line"><a id="l00017" name="l00017"></a><span class="lineno"> 17</span> }</div>
|
||
<div class="line"><a id="l00018" name="l00018"></a><span class="lineno"> 18</span> <span class="keywordtype">void</span> Boost::setHyperparameters(<span class="keyword">const</span> nlohmann::json& hyperparameters_)</div>
|
||
<div class="line"><a id="l00019" name="l00019"></a><span class="lineno"> 19</span> {</div>
|
||
<div class="line"><a id="l00020" name="l00020"></a><span class="lineno"> 20</span> <span class="keyword">auto</span> hyperparameters = hyperparameters_;</div>
|
||
<div class="line"><a id="l00021" name="l00021"></a><span class="lineno"> 21</span> <span class="keywordflow">if</span> (hyperparameters.contains(<span class="stringliteral">"order"</span>)) {</div>
|
||
<div class="line"><a id="l00022" name="l00022"></a><span class="lineno"> 22</span> std::vector<std::string> algos = { Orders.ASC, Orders.DESC, Orders.RAND };</div>
|
||
<div class="line"><a id="l00023" name="l00023"></a><span class="lineno"> 23</span> order_algorithm = hyperparameters[<span class="stringliteral">"order"</span>];</div>
|
||
<div class="line"><a id="l00024" name="l00024"></a><span class="lineno"> 24</span> <span class="keywordflow">if</span> (std::find(algos.begin(), algos.end(), order_algorithm) == algos.end()) {</div>
|
||
<div class="line"><a id="l00025" name="l00025"></a><span class="lineno"> 25</span> <span class="keywordflow">throw</span> std::invalid_argument(<span class="stringliteral">"Invalid order algorithm, valid values ["</span> + Orders.ASC + <span class="stringliteral">", "</span> + Orders.DESC + <span class="stringliteral">", "</span> + Orders.RAND + <span class="stringliteral">"]"</span>);</div>
|
||
<div class="line"><a id="l00026" name="l00026"></a><span class="lineno"> 26</span> }</div>
|
||
<div class="line"><a id="l00027" name="l00027"></a><span class="lineno"> 27</span> hyperparameters.erase(<span class="stringliteral">"order"</span>);</div>
|
||
<div class="line"><a id="l00028" name="l00028"></a><span class="lineno"> 28</span> }</div>
|
||
<div class="line"><a id="l00029" name="l00029"></a><span class="lineno"> 29</span> <span class="keywordflow">if</span> (hyperparameters.contains(<span class="stringliteral">"convergence"</span>)) {</div>
|
||
<div class="line"><a id="l00030" name="l00030"></a><span class="lineno"> 30</span> convergence = hyperparameters[<span class="stringliteral">"convergence"</span>];</div>
|
||
<div class="line"><a id="l00031" name="l00031"></a><span class="lineno"> 31</span> hyperparameters.erase(<span class="stringliteral">"convergence"</span>);</div>
|
||
<div class="line"><a id="l00032" name="l00032"></a><span class="lineno"> 32</span> }</div>
|
||
<div class="line"><a id="l00033" name="l00033"></a><span class="lineno"> 33</span> <span class="keywordflow">if</span> (hyperparameters.contains(<span class="stringliteral">"convergence_best"</span>)) {</div>
|
||
<div class="line"><a id="l00034" name="l00034"></a><span class="lineno"> 34</span> convergence_best = hyperparameters[<span class="stringliteral">"convergence_best"</span>];</div>
|
||
<div class="line"><a id="l00035" name="l00035"></a><span class="lineno"> 35</span> hyperparameters.erase(<span class="stringliteral">"convergence_best"</span>);</div>
|
||
<div class="line"><a id="l00036" name="l00036"></a><span class="lineno"> 36</span> }</div>
|
||
<div class="line"><a id="l00037" name="l00037"></a><span class="lineno"> 37</span> <span class="keywordflow">if</span> (hyperparameters.contains(<span class="stringliteral">"bisection"</span>)) {</div>
|
||
<div class="line"><a id="l00038" name="l00038"></a><span class="lineno"> 38</span> bisection = hyperparameters[<span class="stringliteral">"bisection"</span>];</div>
|
||
<div class="line"><a id="l00039" name="l00039"></a><span class="lineno"> 39</span> hyperparameters.erase(<span class="stringliteral">"bisection"</span>);</div>
|
||
<div class="line"><a id="l00040" name="l00040"></a><span class="lineno"> 40</span> }</div>
|
||
<div class="line"><a id="l00041" name="l00041"></a><span class="lineno"> 41</span> <span class="keywordflow">if</span> (hyperparameters.contains(<span class="stringliteral">"threshold"</span>)) {</div>
|
||
<div class="line"><a id="l00042" name="l00042"></a><span class="lineno"> 42</span> threshold = hyperparameters[<span class="stringliteral">"threshold"</span>];</div>
|
||
<div class="line"><a id="l00043" name="l00043"></a><span class="lineno"> 43</span> hyperparameters.erase(<span class="stringliteral">"threshold"</span>);</div>
|
||
<div class="line"><a id="l00044" name="l00044"></a><span class="lineno"> 44</span> }</div>
|
||
<div class="line"><a id="l00045" name="l00045"></a><span class="lineno"> 45</span> <span class="keywordflow">if</span> (hyperparameters.contains(<span class="stringliteral">"maxTolerance"</span>)) {</div>
|
||
<div class="line"><a id="l00046" name="l00046"></a><span class="lineno"> 46</span> maxTolerance = hyperparameters[<span class="stringliteral">"maxTolerance"</span>];</div>
|
||
<div class="line"><a id="l00047" name="l00047"></a><span class="lineno"> 47</span> <span class="keywordflow">if</span> (maxTolerance < 1 || maxTolerance > 4)</div>
|
||
<div class="line"><a id="l00048" name="l00048"></a><span class="lineno"> 48</span> <span class="keywordflow">throw</span> std::invalid_argument(<span class="stringliteral">"Invalid maxTolerance value, must be greater in [1, 4]"</span>);</div>
|
||
<div class="line"><a id="l00049" name="l00049"></a><span class="lineno"> 49</span> hyperparameters.erase(<span class="stringliteral">"maxTolerance"</span>);</div>
|
||
<div class="line"><a id="l00050" name="l00050"></a><span class="lineno"> 50</span> }</div>
|
||
<div class="line"><a id="l00051" name="l00051"></a><span class="lineno"> 51</span> <span class="keywordflow">if</span> (hyperparameters.contains(<span class="stringliteral">"predict_voting"</span>)) {</div>
|
||
<div class="line"><a id="l00052" name="l00052"></a><span class="lineno"> 52</span> predict_voting = hyperparameters[<span class="stringliteral">"predict_voting"</span>];</div>
|
||
<div class="line"><a id="l00053" name="l00053"></a><span class="lineno"> 53</span> hyperparameters.erase(<span class="stringliteral">"predict_voting"</span>);</div>
|
||
<div class="line"><a id="l00054" name="l00054"></a><span class="lineno"> 54</span> }</div>
|
||
<div class="line"><a id="l00055" name="l00055"></a><span class="lineno"> 55</span> <span class="keywordflow">if</span> (hyperparameters.contains(<span class="stringliteral">"select_features"</span>)) {</div>
|
||
<div class="line"><a id="l00056" name="l00056"></a><span class="lineno"> 56</span> <span class="keyword">auto</span> selectedAlgorithm = hyperparameters[<span class="stringliteral">"select_features"</span>];</div>
|
||
<div class="line"><a id="l00057" name="l00057"></a><span class="lineno"> 57</span> std::vector<std::string> algos = { SelectFeatures.IWSS, SelectFeatures.CFS, SelectFeatures.FCBF };</div>
|
||
<div class="line"><a id="l00058" name="l00058"></a><span class="lineno"> 58</span> selectFeatures = <span class="keyword">true</span>;</div>
|
||
<div class="line"><a id="l00059" name="l00059"></a><span class="lineno"> 59</span> select_features_algorithm = selectedAlgorithm;</div>
|
||
<div class="line"><a id="l00060" name="l00060"></a><span class="lineno"> 60</span> <span class="keywordflow">if</span> (std::find(algos.begin(), algos.end(), selectedAlgorithm) == algos.end()) {</div>
|
||
<div class="line"><a id="l00061" name="l00061"></a><span class="lineno"> 61</span> <span class="keywordflow">throw</span> std::invalid_argument(<span class="stringliteral">"Invalid selectFeatures value, valid values ["</span> + SelectFeatures.IWSS + <span class="stringliteral">", "</span> + SelectFeatures.CFS + <span class="stringliteral">", "</span> + SelectFeatures.FCBF + <span class="stringliteral">"]"</span>);</div>
|
||
<div class="line"><a id="l00062" name="l00062"></a><span class="lineno"> 62</span> }</div>
|
||
<div class="line"><a id="l00063" name="l00063"></a><span class="lineno"> 63</span> hyperparameters.erase(<span class="stringliteral">"select_features"</span>);</div>
|
||
<div class="line"><a id="l00064" name="l00064"></a><span class="lineno"> 64</span> }</div>
|
||
<div class="line"><a id="l00065" name="l00065"></a><span class="lineno"> 65</span> <span class="keywordflow">if</span> (hyperparameters.contains(<span class="stringliteral">"block_update"</span>)) {</div>
|
||
<div class="line"><a id="l00066" name="l00066"></a><span class="lineno"> 66</span> block_update = hyperparameters[<span class="stringliteral">"block_update"</span>];</div>
|
||
<div class="line"><a id="l00067" name="l00067"></a><span class="lineno"> 67</span> hyperparameters.erase(<span class="stringliteral">"block_update"</span>);</div>
|
||
<div class="line"><a id="l00068" name="l00068"></a><span class="lineno"> 68</span> }</div>
|
||
<div class="line"><a id="l00069" name="l00069"></a><span class="lineno"> 69</span> Classifier::setHyperparameters(hyperparameters);</div>
|
||
<div class="line"><a id="l00070" name="l00070"></a><span class="lineno"> 70</span> }</div>
|
||
<div class="line"><a id="l00071" name="l00071"></a><span class="lineno"> 71</span> <span class="keywordtype">void</span> Boost::buildModel(<span class="keyword">const</span> torch::Tensor& weights)</div>
|
||
<div class="line"><a id="l00072" name="l00072"></a><span class="lineno"> 72</span> {</div>
|
||
<div class="line"><a id="l00073" name="l00073"></a><span class="lineno"> 73</span> <span class="comment">// Models shall be built in trainModel</span></div>
|
||
<div class="line"><a id="l00074" name="l00074"></a><span class="lineno"> 74</span> models.clear();</div>
|
||
<div class="line"><a id="l00075" name="l00075"></a><span class="lineno"> 75</span> significanceModels.clear();</div>
|
||
<div class="line"><a id="l00076" name="l00076"></a><span class="lineno"> 76</span> n_models = 0;</div>
|
||
<div class="line"><a id="l00077" name="l00077"></a><span class="lineno"> 77</span> <span class="comment">// Prepare the validation dataset</span></div>
|
||
<div class="line"><a id="l00078" name="l00078"></a><span class="lineno"> 78</span> <span class="keyword">auto</span> y_ = dataset.index({ -1, <span class="stringliteral">"..."</span> });</div>
|
||
<div class="line"><a id="l00079" name="l00079"></a><span class="lineno"> 79</span> <span class="keywordflow">if</span> (convergence) {</div>
|
||
<div class="line"><a id="l00080" name="l00080"></a><span class="lineno"> 80</span> <span class="comment">// Prepare train & validation sets from train data</span></div>
|
||
<div class="line"><a id="l00081" name="l00081"></a><span class="lineno"> 81</span> <span class="keyword">auto</span> fold = folding::StratifiedKFold(5, y_, 271);</div>
|
||
<div class="line"><a id="l00082" name="l00082"></a><span class="lineno"> 82</span> <span class="keyword">auto</span> [train, test] = fold.getFold(0);</div>
|
||
<div class="line"><a id="l00083" name="l00083"></a><span class="lineno"> 83</span> <span class="keyword">auto</span> train_t = torch::tensor(train);</div>
|
||
<div class="line"><a id="l00084" name="l00084"></a><span class="lineno"> 84</span> <span class="keyword">auto</span> test_t = torch::tensor(test);</div>
|
||
<div class="line"><a id="l00085" name="l00085"></a><span class="lineno"> 85</span> <span class="comment">// Get train and validation sets</span></div>
|
||
<div class="line"><a id="l00086" name="l00086"></a><span class="lineno"> 86</span> X_train = dataset.index({ torch::indexing::Slice(0, dataset.size(0) - 1), train_t });</div>
|
||
<div class="line"><a id="l00087" name="l00087"></a><span class="lineno"> 87</span> y_train = dataset.index({ -1, train_t });</div>
|
||
<div class="line"><a id="l00088" name="l00088"></a><span class="lineno"> 88</span> X_test = dataset.index({ torch::indexing::Slice(0, dataset.size(0) - 1), test_t });</div>
|
||
<div class="line"><a id="l00089" name="l00089"></a><span class="lineno"> 89</span> y_test = dataset.index({ -1, test_t });</div>
|
||
<div class="line"><a id="l00090" name="l00090"></a><span class="lineno"> 90</span> dataset = X_train;</div>
|
||
<div class="line"><a id="l00091" name="l00091"></a><span class="lineno"> 91</span> m = X_train.size(1);</div>
|
||
<div class="line"><a id="l00092" name="l00092"></a><span class="lineno"> 92</span> <span class="keyword">auto</span> n_classes = states.at(className).size();</div>
|
||
<div class="line"><a id="l00093" name="l00093"></a><span class="lineno"> 93</span> <span class="comment">// Build dataset with train data</span></div>
|
||
<div class="line"><a id="l00094" name="l00094"></a><span class="lineno"> 94</span> buildDataset(y_train);</div>
|
||
<div class="line"><a id="l00095" name="l00095"></a><span class="lineno"> 95</span> metrics = Metrics(dataset, features, className, n_classes);</div>
|
||
<div class="line"><a id="l00096" name="l00096"></a><span class="lineno"> 96</span> } <span class="keywordflow">else</span> {</div>
|
||
<div class="line"><a id="l00097" name="l00097"></a><span class="lineno"> 97</span> <span class="comment">// Use all data to train</span></div>
|
||
<div class="line"><a id="l00098" name="l00098"></a><span class="lineno"> 98</span> X_train = dataset.index({ torch::indexing::Slice(0, dataset.size(0) - 1), <span class="stringliteral">"..."</span> });</div>
|
||
<div class="line"><a id="l00099" name="l00099"></a><span class="lineno"> 99</span> y_train = y_;</div>
|
||
<div class="line"><a id="l00100" name="l00100"></a><span class="lineno"> 100</span> }</div>
|
||
<div class="line"><a id="l00101" name="l00101"></a><span class="lineno"> 101</span> }</div>
|
||
<div class="line"><a id="l00102" name="l00102"></a><span class="lineno"> 102</span> std::vector<int> Boost::featureSelection(torch::Tensor& weights_)</div>
|
||
<div class="line"><a id="l00103" name="l00103"></a><span class="lineno"> 103</span> {</div>
|
||
<div class="line"><a id="l00104" name="l00104"></a><span class="lineno"> 104</span> <span class="keywordtype">int</span> maxFeatures = 0;</div>
|
||
<div class="line"><a id="l00105" name="l00105"></a><span class="lineno"> 105</span> <span class="keywordflow">if</span> (select_features_algorithm == SelectFeatures.CFS) {</div>
|
||
<div class="line"><a id="l00106" name="l00106"></a><span class="lineno"> 106</span> featureSelector = <span class="keyword">new</span> CFS(dataset, features, className, maxFeatures, states.at(className).size(), weights_);</div>
|
||
<div class="line"><a id="l00107" name="l00107"></a><span class="lineno"> 107</span> } <span class="keywordflow">else</span> <span class="keywordflow">if</span> (select_features_algorithm == SelectFeatures.IWSS) {</div>
|
||
<div class="line"><a id="l00108" name="l00108"></a><span class="lineno"> 108</span> <span class="keywordflow">if</span> (threshold < 0 || threshold >0.5) {</div>
|
||
<div class="line"><a id="l00109" name="l00109"></a><span class="lineno"> 109</span> <span class="keywordflow">throw</span> std::invalid_argument(<span class="stringliteral">"Invalid threshold value for "</span> + SelectFeatures.IWSS + <span class="stringliteral">" [0, 0.5]"</span>);</div>
|
||
<div class="line"><a id="l00110" name="l00110"></a><span class="lineno"> 110</span> }</div>
|
||
<div class="line"><a id="l00111" name="l00111"></a><span class="lineno"> 111</span> featureSelector = <span class="keyword">new</span> IWSS(dataset, features, className, maxFeatures, states.at(className).size(), weights_, threshold);</div>
|
||
<div class="line"><a id="l00112" name="l00112"></a><span class="lineno"> 112</span> } <span class="keywordflow">else</span> <span class="keywordflow">if</span> (select_features_algorithm == SelectFeatures.FCBF) {</div>
|
||
<div class="line"><a id="l00113" name="l00113"></a><span class="lineno"> 113</span> <span class="keywordflow">if</span> (threshold < 1e-7 || threshold > 1) {</div>
|
||
<div class="line"><a id="l00114" name="l00114"></a><span class="lineno"> 114</span> <span class="keywordflow">throw</span> std::invalid_argument(<span class="stringliteral">"Invalid threshold value for "</span> + SelectFeatures.FCBF + <span class="stringliteral">" [1e-7, 1]"</span>);</div>
|
||
<div class="line"><a id="l00115" name="l00115"></a><span class="lineno"> 115</span> }</div>
|
||
<div class="line"><a id="l00116" name="l00116"></a><span class="lineno"> 116</span> featureSelector = <span class="keyword">new</span> FCBF(dataset, features, className, maxFeatures, states.at(className).size(), weights_, threshold);</div>
|
||
<div class="line"><a id="l00117" name="l00117"></a><span class="lineno"> 117</span> }</div>
|
||
<div class="line"><a id="l00118" name="l00118"></a><span class="lineno"> 118</span> featureSelector->fit();</div>
|
||
<div class="line"><a id="l00119" name="l00119"></a><span class="lineno"> 119</span> <span class="keyword">auto</span> featuresUsed = featureSelector->getFeatures();</div>
|
||
<div class="line"><a id="l00120" name="l00120"></a><span class="lineno"> 120</span> <span class="keyword">delete</span> featureSelector;</div>
|
||
<div class="line"><a id="l00121" name="l00121"></a><span class="lineno"> 121</span> <span class="keywordflow">return</span> featuresUsed;</div>
|
||
<div class="line"><a id="l00122" name="l00122"></a><span class="lineno"> 122</span> }</div>
|
||
<div class="line"><a id="l00123" name="l00123"></a><span class="lineno"> 123</span> std::tuple<torch::Tensor&, double, bool> Boost::update_weights(torch::Tensor& ytrain, torch::Tensor& ypred, torch::Tensor& weights)</div>
|
||
<div class="line"><a id="l00124" name="l00124"></a><span class="lineno"> 124</span> {</div>
|
||
<div class="line"><a id="l00125" name="l00125"></a><span class="lineno"> 125</span> <span class="keywordtype">bool</span> terminate = <span class="keyword">false</span>;</div>
|
||
<div class="line"><a id="l00126" name="l00126"></a><span class="lineno"> 126</span> <span class="keywordtype">double</span> alpha_t = 0;</div>
|
||
<div class="line"><a id="l00127" name="l00127"></a><span class="lineno"> 127</span> <span class="keyword">auto</span> mask_wrong = ypred != ytrain;</div>
|
||
<div class="line"><a id="l00128" name="l00128"></a><span class="lineno"> 128</span> <span class="keyword">auto</span> mask_right = ypred == ytrain;</div>
|
||
<div class="line"><a id="l00129" name="l00129"></a><span class="lineno"> 129</span> <span class="keyword">auto</span> masked_weights = weights * mask_wrong.to(weights.dtype());</div>
|
||
<div class="line"><a id="l00130" name="l00130"></a><span class="lineno"> 130</span> <span class="keywordtype">double</span> epsilon_t = masked_weights.sum().item<<span class="keywordtype">double</span>>();</div>
|
||
<div class="line"><a id="l00131" name="l00131"></a><span class="lineno"> 131</span> <span class="keywordflow">if</span> (epsilon_t > 0.5) {</div>
|
||
<div class="line"><a id="l00132" name="l00132"></a><span class="lineno"> 132</span> <span class="comment">// Inverse the weights policy (plot ln(wt))</span></div>
|
||
<div class="line"><a id="l00133" name="l00133"></a><span class="lineno"> 133</span> <span class="comment">// "In each round of AdaBoost, there is a sanity check to ensure that the current base </span></div>
|
||
<div class="line"><a id="l00134" name="l00134"></a><span class="lineno"> 134</span> <span class="comment">// learner is better than random guess" (Zhi-Hua Zhou, 2012)</span></div>
|
||
<div class="line"><a id="l00135" name="l00135"></a><span class="lineno"> 135</span> terminate = <span class="keyword">true</span>;</div>
|
||
<div class="line"><a id="l00136" name="l00136"></a><span class="lineno"> 136</span> } <span class="keywordflow">else</span> {</div>
|
||
<div class="line"><a id="l00137" name="l00137"></a><span class="lineno"> 137</span> <span class="keywordtype">double</span> wt = (1 - epsilon_t) / epsilon_t;</div>
|
||
<div class="line"><a id="l00138" name="l00138"></a><span class="lineno"> 138</span> alpha_t = epsilon_t == 0 ? 1 : 0.5 * log(wt);</div>
|
||
<div class="line"><a id="l00139" name="l00139"></a><span class="lineno"> 139</span> <span class="comment">// Step 3.2: Update weights for next classifier</span></div>
|
||
<div class="line"><a id="l00140" name="l00140"></a><span class="lineno"> 140</span> <span class="comment">// Step 3.2.1: Update weights of wrong samples</span></div>
|
||
<div class="line"><a id="l00141" name="l00141"></a><span class="lineno"> 141</span> weights += mask_wrong.to(weights.dtype()) * exp(alpha_t) * weights;</div>
|
||
<div class="line"><a id="l00142" name="l00142"></a><span class="lineno"> 142</span> <span class="comment">// Step 3.2.2: Update weights of right samples</span></div>
|
||
<div class="line"><a id="l00143" name="l00143"></a><span class="lineno"> 143</span> weights += mask_right.to(weights.dtype()) * exp(-alpha_t) * weights;</div>
|
||
<div class="line"><a id="l00144" name="l00144"></a><span class="lineno"> 144</span> <span class="comment">// Step 3.3: Normalise the weights</span></div>
|
||
<div class="line"><a id="l00145" name="l00145"></a><span class="lineno"> 145</span> <span class="keywordtype">double</span> totalWeights = torch::sum(weights).item<<span class="keywordtype">double</span>>();</div>
|
||
<div class="line"><a id="l00146" name="l00146"></a><span class="lineno"> 146</span> weights = weights / totalWeights;</div>
|
||
<div class="line"><a id="l00147" name="l00147"></a><span class="lineno"> 147</span> }</div>
|
||
<div class="line"><a id="l00148" name="l00148"></a><span class="lineno"> 148</span> <span class="keywordflow">return</span> { weights, alpha_t, terminate };</div>
|
||
<div class="line"><a id="l00149" name="l00149"></a><span class="lineno"> 149</span> }</div>
|
||
<div class="line"><a id="l00150" name="l00150"></a><span class="lineno"> 150</span> std::tuple<torch::Tensor&, double, bool> Boost::update_weights_block(<span class="keywordtype">int</span> k, torch::Tensor& ytrain, torch::Tensor& weights)</div>
|
||
<div class="line"><a id="l00151" name="l00151"></a><span class="lineno"> 151</span> {</div>
|
||
<div class="line"><a id="l00152" name="l00152"></a><span class="lineno"> 152</span> <span class="comment">/* Update Block algorithm</span></div>
|
||
<div class="line"><a id="l00153" name="l00153"></a><span class="lineno"> 153</span><span class="comment"> k = # of models in block</span></div>
|
||
<div class="line"><a id="l00154" name="l00154"></a><span class="lineno"> 154</span><span class="comment"> n_models = # of models in ensemble to make predictions</span></div>
|
||
<div class="line"><a id="l00155" name="l00155"></a><span class="lineno"> 155</span><span class="comment"> n_models_bak = # models saved</span></div>
|
||
<div class="line"><a id="l00156" name="l00156"></a><span class="lineno"> 156</span><span class="comment"> models = vector of models to make predictions</span></div>
|
||
<div class="line"><a id="l00157" name="l00157"></a><span class="lineno"> 157</span><span class="comment"> models_bak = models not used to make predictions</span></div>
|
||
<div class="line"><a id="l00158" name="l00158"></a><span class="lineno"> 158</span><span class="comment"> significances_bak = backup of significances vector</span></div>
|
||
<div class="line"><a id="l00159" name="l00159"></a><span class="lineno"> 159</span><span class="comment"></span> </div>
|
||
<div class="line"><a id="l00160" name="l00160"></a><span class="lineno"> 160</span><span class="comment"> Case list</span></div>
|
||
<div class="line"><a id="l00161" name="l00161"></a><span class="lineno"> 161</span><span class="comment"> A) k = 1, n_models = 1 => n = 0 , n_models = n + k</span></div>
|
||
<div class="line"><a id="l00162" name="l00162"></a><span class="lineno"> 162</span><span class="comment"> B) k = 1, n_models = n + 1 => n_models = n + k</span></div>
|
||
<div class="line"><a id="l00163" name="l00163"></a><span class="lineno"> 163</span><span class="comment"> C) k > 1, n_models = k + 1 => n= 1, n_models = n + k</span></div>
|
||
<div class="line"><a id="l00164" name="l00164"></a><span class="lineno"> 164</span><span class="comment"> D) k > 1, n_models = k => n = 0, n_models = n + k</span></div>
|
||
<div class="line"><a id="l00165" name="l00165"></a><span class="lineno"> 165</span><span class="comment"> E) k > 1, n_models = k + n => n_models = n + k</span></div>
|
||
<div class="line"><a id="l00166" name="l00166"></a><span class="lineno"> 166</span><span class="comment"></span> </div>
|
||
<div class="line"><a id="l00167" name="l00167"></a><span class="lineno"> 167</span><span class="comment"> A, D) n=0, k > 0, n_models == k</span></div>
|
||
<div class="line"><a id="l00168" name="l00168"></a><span class="lineno"> 168</span><span class="comment"> 1. n_models_bak <- n_models</span></div>
|
||
<div class="line"><a id="l00169" name="l00169"></a><span class="lineno"> 169</span><span class="comment"> 2. significances_bak <- significances</span></div>
|
||
<div class="line"><a id="l00170" name="l00170"></a><span class="lineno"> 170</span><span class="comment"> 3. significances = vector(k, 1)</span></div>
|
||
<div class="line"><a id="l00171" name="l00171"></a><span class="lineno"> 171</span><span class="comment"> 4. Don’t move any classifiers out of models</span></div>
|
||
<div class="line"><a id="l00172" name="l00172"></a><span class="lineno"> 172</span><span class="comment"> 5. n_models <- k</span></div>
|
||
<div class="line"><a id="l00173" name="l00173"></a><span class="lineno"> 173</span><span class="comment"> 6. Make prediction, compute alpha, update weights</span></div>
|
||
<div class="line"><a id="l00174" name="l00174"></a><span class="lineno"> 174</span><span class="comment"> 7. Don’t restore any classifiers to models</span></div>
|
||
<div class="line"><a id="l00175" name="l00175"></a><span class="lineno"> 175</span><span class="comment"> 8. significances <- significances_bak</span></div>
|
||
<div class="line"><a id="l00176" name="l00176"></a><span class="lineno"> 176</span><span class="comment"> 9. Update last k significances</span></div>
|
||
<div class="line"><a id="l00177" name="l00177"></a><span class="lineno"> 177</span><span class="comment"> 10. n_models <- n_models_bak</span></div>
|
||
<div class="line"><a id="l00178" name="l00178"></a><span class="lineno"> 178</span><span class="comment"></span> </div>
|
||
<div class="line"><a id="l00179" name="l00179"></a><span class="lineno"> 179</span><span class="comment"> B, C, E) n > 0, k > 0, n_models == n + k</span></div>
|
||
<div class="line"><a id="l00180" name="l00180"></a><span class="lineno"> 180</span><span class="comment"> 1. n_models_bak <- n_models</span></div>
|
||
<div class="line"><a id="l00181" name="l00181"></a><span class="lineno"> 181</span><span class="comment"> 2. significances_bak <- significances</span></div>
|
||
<div class="line"><a id="l00182" name="l00182"></a><span class="lineno"> 182</span><span class="comment"> 3. significances = vector(k, 1)</span></div>
|
||
<div class="line"><a id="l00183" name="l00183"></a><span class="lineno"> 183</span><span class="comment"> 4. Move first n classifiers to models_bak</span></div>
|
||
<div class="line"><a id="l00184" name="l00184"></a><span class="lineno"> 184</span><span class="comment"> 5. n_models <- k</span></div>
|
||
<div class="line"><a id="l00185" name="l00185"></a><span class="lineno"> 185</span><span class="comment"> 6. Make prediction, compute alpha, update weights</span></div>
|
||
<div class="line"><a id="l00186" name="l00186"></a><span class="lineno"> 186</span><span class="comment"> 7. Insert classifiers in models_bak to be the first n models</span></div>
|
||
<div class="line"><a id="l00187" name="l00187"></a><span class="lineno"> 187</span><span class="comment"> 8. significances <- significances_bak</span></div>
|
||
<div class="line"><a id="l00188" name="l00188"></a><span class="lineno"> 188</span><span class="comment"> 9. Update last k significances</span></div>
|
||
<div class="line"><a id="l00189" name="l00189"></a><span class="lineno"> 189</span><span class="comment"> 10. n_models <- n_models_bak</span></div>
|
||
<div class="line"><a id="l00190" name="l00190"></a><span class="lineno"> 190</span><span class="comment"> */</span></div>
|
||
<div class="line"><a id="l00191" name="l00191"></a><span class="lineno"> 191</span> <span class="comment">//</span></div>
|
||
<div class="line"><a id="l00192" name="l00192"></a><span class="lineno"> 192</span> <span class="comment">// Make predict with only the last k models</span></div>
|
||
<div class="line"><a id="l00193" name="l00193"></a><span class="lineno"> 193</span> <span class="comment">//</span></div>
|
||
<div class="line"><a id="l00194" name="l00194"></a><span class="lineno"> 194</span> std::unique_ptr<Classifier> model;</div>
|
||
<div class="line"><a id="l00195" name="l00195"></a><span class="lineno"> 195</span> std::vector<std::unique_ptr<Classifier>> models_bak;</div>
|
||
<div class="line"><a id="l00196" name="l00196"></a><span class="lineno"> 196</span> <span class="comment">// 1. n_models_bak <- n_models 2. significances_bak <- significances</span></div>
|
||
<div class="line"><a id="l00197" name="l00197"></a><span class="lineno"> 197</span> <span class="keyword">auto</span> significance_bak = significanceModels;</div>
|
||
<div class="line"><a id="l00198" name="l00198"></a><span class="lineno"> 198</span> <span class="keyword">auto</span> n_models_bak = n_models;</div>
|
||
<div class="line"><a id="l00199" name="l00199"></a><span class="lineno"> 199</span> <span class="comment">// 3. significances = vector(k, 1)</span></div>
|
||
<div class="line"><a id="l00200" name="l00200"></a><span class="lineno"> 200</span> significanceModels = std::vector<double>(k, 1.0);</div>
|
||
<div class="line"><a id="l00201" name="l00201"></a><span class="lineno"> 201</span> <span class="comment">// 4. Move first n classifiers to models_bak</span></div>
|
||
<div class="line"><a id="l00202" name="l00202"></a><span class="lineno"> 202</span> <span class="comment">// backup the first n_models - k models (if n_models == k, don't backup any)</span></div>
|
||
<div class="line"><a id="l00203" name="l00203"></a><span class="lineno"> 203</span> <span class="keywordflow">for</span> (<span class="keywordtype">int</span> i = 0; i < n_models - k; ++i) {</div>
|
||
<div class="line"><a id="l00204" name="l00204"></a><span class="lineno"> 204</span> model = std::move(models[0]);</div>
|
||
<div class="line"><a id="l00205" name="l00205"></a><span class="lineno"> 205</span> models.erase(models.begin());</div>
|
||
<div class="line"><a id="l00206" name="l00206"></a><span class="lineno"> 206</span> models_bak.push_back(std::move(model));</div>
|
||
<div class="line"><a id="l00207" name="l00207"></a><span class="lineno"> 207</span> }</div>
|
||
<div class="line"><a id="l00208" name="l00208"></a><span class="lineno"> 208</span> assert(models.size() == k);</div>
|
||
<div class="line"><a id="l00209" name="l00209"></a><span class="lineno"> 209</span> <span class="comment">// 5. n_models <- k</span></div>
|
||
<div class="line"><a id="l00210" name="l00210"></a><span class="lineno"> 210</span> n_models = k;</div>
|
||
<div class="line"><a id="l00211" name="l00211"></a><span class="lineno"> 211</span> <span class="comment">// 6. Make prediction, compute alpha, update weights</span></div>
|
||
<div class="line"><a id="l00212" name="l00212"></a><span class="lineno"> 212</span> <span class="keyword">auto</span> ypred = predict(X_train);</div>
|
||
<div class="line"><a id="l00213" name="l00213"></a><span class="lineno"> 213</span> <span class="comment">//</span></div>
|
||
<div class="line"><a id="l00214" name="l00214"></a><span class="lineno"> 214</span> <span class="comment">// Update weights</span></div>
|
||
<div class="line"><a id="l00215" name="l00215"></a><span class="lineno"> 215</span> <span class="comment">//</span></div>
|
||
<div class="line"><a id="l00216" name="l00216"></a><span class="lineno"> 216</span> <span class="keywordtype">double</span> alpha_t;</div>
|
||
<div class="line"><a id="l00217" name="l00217"></a><span class="lineno"> 217</span> <span class="keywordtype">bool</span> terminate;</div>
|
||
<div class="line"><a id="l00218" name="l00218"></a><span class="lineno"> 218</span> std::tie(weights, alpha_t, terminate) = update_weights(y_train, ypred, weights);</div>
|
||
<div class="line"><a id="l00219" name="l00219"></a><span class="lineno"> 219</span> <span class="comment">//</span></div>
|
||
<div class="line"><a id="l00220" name="l00220"></a><span class="lineno"> 220</span> <span class="comment">// Restore the models if needed</span></div>
|
||
<div class="line"><a id="l00221" name="l00221"></a><span class="lineno"> 221</span> <span class="comment">//</span></div>
|
||
<div class="line"><a id="l00222" name="l00222"></a><span class="lineno"> 222</span> <span class="comment">// 7. Insert classifiers in models_bak to be the first n models</span></div>
|
||
<div class="line"><a id="l00223" name="l00223"></a><span class="lineno"> 223</span> <span class="comment">// if n_models_bak == k, don't restore any, because none of them were moved</span></div>
|
||
<div class="line"><a id="l00224" name="l00224"></a><span class="lineno"> 224</span> <span class="keywordflow">if</span> (k != n_models_bak) {</div>
|
||
<div class="line"><a id="l00225" name="l00225"></a><span class="lineno"> 225</span> <span class="comment">// Insert in the same order as they were extracted</span></div>
|
||
<div class="line"><a id="l00226" name="l00226"></a><span class="lineno"> 226</span> <span class="keywordtype">int</span> bak_size = models_bak.size();</div>
|
||
<div class="line"><a id="l00227" name="l00227"></a><span class="lineno"> 227</span> <span class="keywordflow">for</span> (<span class="keywordtype">int</span> i = 0; i < bak_size; ++i) {</div>
|
||
<div class="line"><a id="l00228" name="l00228"></a><span class="lineno"> 228</span> model = std::move(models_bak[bak_size - 1 - i]);</div>
|
||
<div class="line"><a id="l00229" name="l00229"></a><span class="lineno"> 229</span> models_bak.erase(models_bak.end() - 1);</div>
|
||
<div class="line"><a id="l00230" name="l00230"></a><span class="lineno"> 230</span> models.insert(models.begin(), std::move(model));</div>
|
||
<div class="line"><a id="l00231" name="l00231"></a><span class="lineno"> 231</span> }</div>
|
||
<div class="line"><a id="l00232" name="l00232"></a><span class="lineno"> 232</span> }</div>
|
||
<div class="line"><a id="l00233" name="l00233"></a><span class="lineno"> 233</span> <span class="comment">// 8. significances <- significances_bak</span></div>
|
||
<div class="line"><a id="l00234" name="l00234"></a><span class="lineno"> 234</span> significanceModels = significance_bak;</div>
|
||
<div class="line"><a id="l00235" name="l00235"></a><span class="lineno"> 235</span> <span class="comment">//</span></div>
|
||
<div class="line"><a id="l00236" name="l00236"></a><span class="lineno"> 236</span> <span class="comment">// Update the significance of the last k models</span></div>
|
||
<div class="line"><a id="l00237" name="l00237"></a><span class="lineno"> 237</span> <span class="comment">//</span></div>
|
||
<div class="line"><a id="l00238" name="l00238"></a><span class="lineno"> 238</span> <span class="comment">// 9. Update last k significances</span></div>
|
||
<div class="line"><a id="l00239" name="l00239"></a><span class="lineno"> 239</span> <span class="keywordflow">for</span> (<span class="keywordtype">int</span> i = 0; i < k; ++i) {</div>
|
||
<div class="line"><a id="l00240" name="l00240"></a><span class="lineno"> 240</span> significanceModels[n_models_bak - k + i] = alpha_t;</div>
|
||
<div class="line"><a id="l00241" name="l00241"></a><span class="lineno"> 241</span> }</div>
|
||
<div class="line"><a id="l00242" name="l00242"></a><span class="lineno"> 242</span> <span class="comment">// 10. n_models <- n_models_bak</span></div>
|
||
<div class="line"><a id="l00243" name="l00243"></a><span class="lineno"> 243</span> n_models = n_models_bak;</div>
|
||
<div class="line"><a id="l00244" name="l00244"></a><span class="lineno"> 244</span> <span class="keywordflow">return</span> { weights, alpha_t, terminate };</div>
|
||
<div class="line"><a id="l00245" name="l00245"></a><span class="lineno"> 245</span> }</div>
|
||
<div class="line"><a id="l00246" name="l00246"></a><span class="lineno"> 246</span>}</div>
|
||
</div><!-- fragment --></div><!-- contents -->
|
||
</div><!-- doc-content -->
|
||
<!-- start footer part -->
|
||
<div id="nav-path" class="navpath"><!-- id is needed for treeview function! -->
|
||
<ul>
|
||
<li class="navelem"><a class="el" href="dir_40070fdff85d618b4d1d3ab4ac4f79bb.html">bayesnet</a></li><li class="navelem"><a class="el" href="dir_2f68445c4ac4316280c650d0a13b2741.html">ensembles</a></li><li class="navelem"><b>Boost.cc</b></li>
|
||
<li class="footer">Generated by <a href="https://www.doxygen.org/index.html"><img class="footer" src="doxygen.svg" width="104" height="31" alt="doxygen"/></a> 1.11.0 </li>
|
||
</ul>
|
||
</div>
|
||
</body>
|
||
</html>
|