<?xml version="1.0"?> 
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Strict//EN" "mathml.dtd"> 
<?xml-stylesheet type="text/css" href="thesis.css"?> 
<html  
xmlns="http://www.w3.org/1999/xhtml"  
><head><title>16.1 Test Error Bound Calculation</title> 
<meta http-equiv="Content-Type" content="text/html; charset=iso-8859-1" /> 
<meta name="generator" content="TeX4ht (http://www.cis.ohio-state.edu/~gurari/TeX4ht/mn.html)" /> 
<meta name="originator" content="TeX4ht (http://www.cis.ohio-state.edu/~gurari/TeX4ht/mn.html)" /> 
<!-- 3,early_,early^,xhtml,mozilla --> 
<meta name="src" content="thesis.tex" /> 
<meta name="date" content="2002-08-28 13:56:00" /> 
<link rel="stylesheet" type="text/css" href="thesis.css" /> 
</head><body 
>
   <div class="crosslinks"><p class="noindent">[<a 
href="thesisse61.xml" >next</a>] [<a 
href="#tailthesisse60.xml">tail</a>] [<a 
href="thesisch16.xml#thesisse60.xml" >up</a>] </p></div>
   <h3 class="sectionHead"><span class="titlemark">16.1. </span> <a 
  name="x86-12900016.1"></a>Test Error Bound Calculation</h3>
<!--l. 5591--><p class="noindent">We can use the holdout bound to calculate upper and lower true error bounds with the
program &#x201C;bound&#x201D;. Here is an example of the application.
                                                                     

                                                                     
</p>
   <table width="100%" 
class="verbatim"><tr class="verbatim"><td 
class="verbatim"><div class="verbatim">
11:32PM&#x00A0;z-12:&#x00A0;echo&#x00A0;&#x0022;test_examples&#x00A0;600
&#x00A0;<br />test_errors&#x00A0;192
&#x00A0;<br />delta&#x00A0;0.025&#x0022;&#x00A0;|&#x00A0;bound
&#x00A0;<br />Applying&#x00A0;varying&#x00A0;approximation&#x00A0;tail&#x00A0;bound
&#x00A0;<br />delta&#x00A0;0.025
&#x00A0;<br />lower_delta&#x00A0;0.5
&#x00A0;<br />test_examples&#x00A0;600
&#x00A0;<br />test_errors&#x00A0;192
&#x00A0;<br />approximation&#x00A0;automatic
&#x00A0;<br />true_error&#x00A0;=&#x00A0;0.32&#x00A0;0.358976827934&#x00A0;0.31926717516
</div>
</td></tr></table>
<!--l. 5608--><p class="indent">   There are several arguments passed to the program. These include:
</p><!--l. 5610--><p class="indent">
           </p><ol type="1" class="enumerate1" start="1" 
>
        <li class="enumerate"><a 
  name="x86-129002x1"></a>test_examples &#x003C;int&#x003E;: the number of test examples.
           </li>
        <li class="enumerate"><a 
  name="x86-129004x2"></a>test_errors &#x003C;int&#x003E;: the number of test errors.
           </li>
        <li class="enumerate"><a 
  name="x86-129006x3"></a>delta &#x003C;float&#x003E;: the probability of failure of the upper bound.
           </li>
        <li class="enumerate"><a 
  name="x86-129008x4"></a>lower_delta &#x003C;float&#x003E;: the probability of failure of the lower bound.</li></ol>
<!--l. 5615--><p class="nopar"> The output of the program is several lines stating the used and assumed arguments and
a final line of the form:
                                                                     

                                                                     
</p>
   <table width="100%" 
class="verbatim"><tr class="verbatim"><td 
class="verbatim"><div class="verbatim">
true_error&#x00A0;=&#x00A0;&#x003C;error&#x00A0;rate&#x003E;&#x00A0;&#x003C;upper_bound&#x003E;&#x00A0;&#x003C;lower_bound&#x003E;
</div>
</td></tr></table>
<!--l. 5623--><p class="indent">   The output of this particular application implies that the true error rate of the chosen hypothesis
is less than <!--l. 5624--><math 
xmlns="http://www.w3.org/1998/Math/MathML" 
mode="inline">        <mrow 
><mn>0</mn><mo 
class="MathClass-punc">.</mo><mn>3</mn><mn>5</mn><mn>8</mn><mn>9</mn><mn>7</mn></mrow></math>
with high confidence (over the drawn sample set).
</p><!--l. 5627--><p class="indent">   For simplicity, the arguments can also be placed into a file:
                                                                     

                                                                     
</p>
   <table width="100%" 
class="verbatim"><tr class="verbatim"><td 
class="verbatim"><div class="verbatim">
11:35PM&#x00A0;z-14:&#x00A0;cat&#x00A0;test_error
&#x00A0;<br />test_examples&#x00A0;800
&#x00A0;<br />test_errors&#x00A0;192
&#x00A0;<br />delta&#x00A0;0.025
&#x00A0;<br />lower_delta&#x00A0;0.025
&#x00A0;<br />11:35PM&#x00A0;z-15:&#x00A0;bound&#x00A0;test_error
&#x00A0;<br />Applying&#x00A0;varying&#x00A0;approximation&#x00A0;tail&#x00A0;bound
&#x00A0;<br />delta&#x00A0;0.025
&#x00A0;<br />lower_delta&#x00A0;0.025
&#x00A0;<br />test_examples&#x00A0;800
&#x00A0;<br />test_errors&#x00A0;192
&#x00A0;<br />approximation&#x00A0;automatic
&#x00A0;<br />true_error&#x00A0;=&#x00A0;0.24&#x00A0;0.28252880089&#x00A0;0.20065891277
</div>
</td></tr></table>
<!--l. 5645--><p class="indent">   The output of this program can be interpreted as a confidence interval.
</p><!--l. 5648--><p class="indent">
                                                                     

                                                                     
</p>
   <div class="crosslinks"><p class="noindent">[<a 
href="thesisse61.xml" >next</a>] [<a 
href="thesisse60.xml" >front</a>] [<a 
href="thesisch16.xml#thesisse60.xml" >up</a>] </p></div><a 
  name="tailthesisse60.xml"></a>  
</body> 
</html> 
