tutorial.xml

使用具有增量学习的监控式学习方法。包括几个不同的分类算法。

<chapter id="tutorial">
<title>Tutorial</title>

<para>
This short tutorial shows how to use <application>Select</application> for doing some email classification testing.
It does not show, however, how to practically use it for classifying your incoming email.
In fact, the current version of <application>Select</application> is not very suitable for such purposes.
I encourage everyone who is interested, though, to look at <xref linkend="select_ifile"/> and <xref linkend="natural"/>, and provide me with feedback of how to improve <application>Select</application> in this area.
</para>

<sect1>
<title>Preparations</title>

<para>
First you need to compile <application>Select</application>, see <xref linkend="install"/>.
Note, however, that this tutorial has been made to work without needing to actually install <application>Select</application>. So it is not necessary to do <userinput>make install</userinput>.
</para>

<para>
For the rest of this tutorial you are supposed to be in the directory <filename>tutorial</filename>.
<screen><prompt>$</prompt> <userinput>cd tutorial</userinput>
</screen>
</para>

<para>
You also need to get some data to use for training and testing.
In this tutorial a subset of the <ulink url="http://kdd.ics.uci.edu/databases/20newsgroups/20newsgroups.html">20-Newsgroups</ulink> collection will be used.
This can be downloaded from <ulink url="http://kdd.ics.uci.edu/databases/20newsgroups/mini_newsgroups.tar.gz">here</ulink> (1.8MB).
Unpack this file like this:
<screen><prompt>$</prompt> <userinput>gzip -cd mini_newsgroups.tar.gz | tar x</userinput></screen>
</para>
</sect1>

<sect1>
<title>Running selectd</title>

<para>
<application>Select</application> uses a separate process to do all classification.
This is started by running <application>selectd</application>, the Select daemon.
<application>selectd</application> is configured using a configuration file. By default it is assumed that this file is called <filename>selectd.conf</filename> and is located in a directory <filename>.select</filename> in your home directory. The <option>-f</option> switch can be used to select another configuration file.
See <xref linkend="selectd"/> for more information about this.
<screen><prompt>$</prompt> <userinput>../selectd -f selectd.conf &amp;</userinput>
</screen>
</para>
</sect1>

<sect1>
<title>Running select_test</title>

<para>
Run <application>select_test</application> and redirect the output into a file.
<screen><prompt>$</prompt> <userinput>../select_test -f select_test.conf &gt; TEST</userinput>
</screen>
</para>

<para>
Now you can kill <application>selectd</application>.
<screen><prompt>$</prompt> <userinput>ps | grep selectd</userinput>
 4711 pts/0    00:00:10 selectd
<prompt>$</prompt> <userinput>kill 4711</userinput>
</screen>
</para>

<para>
Now view the last lines of the file.
You can do this with the command <command>tail</command>:
<screen><prompt>$</prompt> <userinput>tail -n 12 TEST</userinput>
<computeroutput>## Classifier 0 (plotted)
##   Correct: 1383  Top-3-Correct: 1750  Covered: 1999  Total: 2000
##   Accuracy: 0.692  Top-3-Accuracy: 0.875  Coverage: 1.000
##   Total-Accuracy: 0.692  Total-Top-3-Accuracy: 0.875
## Classifier 1
##   Correct: 129  Top-3-Correct: 129  Covered: 169  Total: 2000
##   Accuracy: 0.763  Top-3-Accuracy: 0.763  Coverage: 0.085
##   Total-Accuracy: 0.065  Total-Top-3-Accuracy: 0.065
## Classifier 2
##   Correct: 178  Top-3-Correct: 178  Covered: 200  Total: 2000
##   Accuracy: 0.890  Top-3-Accuracy: 0.890  Coverage: 0.100
##   Total-Accuracy: 0.089  Total-Top-3-Accuracy: 0.089
</computeroutput></screen>
</para>

<sect2>
<title>Making a Plot</title>

<para>
If you have the application <application>gnuplot</application> installed, then you can make a plot of the test run.
<screen><prompt>$</prompt> <userinput>gnuplot</userinput>
<prompt>gnuplot></prompt> <userinput>plot "TEST" notitle with lines</userinput>
</screen>
<figure><title>Test plot</title>
<mediaobject>
<imageobject><imagedata fileref="test.png" format="PNG"/></imageobject>
<textobject><phrase>Test plot</phrase></textobject>
</mediaobject>
</figure>
The plot shows how the classification performance varies over time.
</para>
</sect2>

</sect1>

</chapter>