📄 page_96.html
字号:
<HTML> <HEAD> <!--SCRIPT LANGUAGE="JavaScript" SRC="http://a1835.g.akamai.net/f/1835/276/3h/www.netlibrary.com/include/js/dictionary_library.js"></SCRIPT> <SCRIPT LANGUAGE="JavaScript"> if (!opener){document.onkeyup=parent.turnBookPage;} </SCRIPT!--> <META HTTP-EQUIV="Cache-Control" CONTENT="no-cache"> <META HTTP-EQUIV="Pragma" CONTENT="no-cache"> <META HTTP-EQUIV="Expires" CONTENT="-1"><META http-equiv="Content-Type" content="text/html; charset=windows-1252"><SCRIPT>var PrevPage="Page_95";var NextPage="Page_97";var CurPage="Page_96";var PageOrder="107";</SCRIPT> <TITLE>Document</TITLE> </HEAD> <BODY BGCOLOR="#FFFFFF"><CENTER><TABLE BORDER=0 WIDTH=100% CELLPADDING=0><TR><TD ALIGN=CENTER> <TABLE BORDER=0 CELLPADDING=2 CELLSPACING=0 WIDTH=100%> <TR> <TD ALIGN=LEFT><A HREF='Page_95.html'>Previous</A></TD> <TD ALIGN=RIGHT><A HREF='Page_97.html'>Next</A></TD> </TR> </TABLE></TD></TR><TR><TD ALIGN=LEFT><P><A NAME='JUMPDEST_Page_96'/><A NAME='{34E}'/><TABLE BORDER=0 CELLSPACING=0 CELLPADDING=0 WIDTH='100%'><TR><TD ALIGN=RIGHT><FONT FACE='Times New Roman, Times, Serif' SIZE=2 COLOR=#FF0000>Page 96</FONT></TD></TR></TABLE><A NAME='{34F}'/><TABLE BORDER=0 CELLSPACING=0 CELLPADDING=0><TR> <TD ROWSPAN=5></TD> <TD COLSPAN=3 HEIGHT=12></TD> <TD ROWSPAN=5></TD></TR><TR> <TD COLSPAN=3></TD></TR><TR><TD></TD> <TD><FONT FACE='Times New Roman, Times, Serif' SIZE=3>The key difference between back-propagation nets and Kohonen nets is that the former is used for supervised learning where a training sample is needed to construct a classification model. For example, with a back-propagation neural network, training is usually done with a balanced sample of website visitors who made a purchase versus those who did not. Also unlike the back-propagation networks which usually have a single output, an SOM network can have multiple outputs competing with each other. SOMs involve unsupervised learning, a training sample for which no output is known, and is commonly used to discover relations in a data set. This by the way does not preclude using both paradigms—first the SOM to discover a sub-class within the data, then a back-propagation network for classification of that class. The same can be said for using neural networks in combination with genetic and machine-learning algorithms.</FONT></TD><TD></TD></TR><TR> <TD COLSPAN=3></TD></TR><TR> <TD COLSPAN=3 HEIGHT=1></TD></TR></TABLE><A NAME='{350}'/><TABLE BORDER=0 CELLSPACING=0 CELLPADDING=0><TR> <TD ROWSPAN=5></TD> <TD COLSPAN=3 HEIGHT=12></TD> <TD ROWSPAN=5></TD></TR><TR> <TD COLSPAN=3></TD></TR><TR><TD></TD> <TD><FONT FACE='Times New Roman, Times, Serif' SIZE=3>SOMs are probably second only to back-propagation architecture in terms of the number of applications and tools for which it is being used. The most significant difference between the back-propagation and the self-organization models is the fact that the SOM is trained without supervision. The SOM network is more biologically oriented than the back-propagation models in that it bears more resemblance to the way humans learn and the brain is organized. If you don't really know what you are attempting to classify (such as the website patterns you are looking for), or if you feel there may be more than one way to categorize your visitors or customers, you may want to start with an SOM.</FONT></TD><TD></TD></TR><TR> <TD COLSPAN=3></TD></TR><TR> <TD COLSPAN=3 HEIGHT=1></TD></TR></TABLE><A NAME='{351}'/><TABLE BORDER=0 CELLSPACING=0 CELLPADDING=0><TR> <TD ROWSPAN=5></TD> <TD COLSPAN=3 HEIGHT=17></TD> <TD ROWSPAN=5></TD></TR><TR> <TD COLSPAN=3></TD></TR><TR><TD></TD> <TD><FONT FACE='Times New Roman, Times, Serif' SIZE=3><B>Calibrating Clusters</B></FONT></TD><TD></TD></TR><TR> <TD COLSPAN=3></TD></TR><TR> <TD COLSPAN=3 HEIGHT=1></TD></TR></TABLE><A NAME='{352}'/><TABLE BORDER=0 CELLSPACING=0 CELLPADDING=0><TR> <TD ROWSPAN=5></TD> <TD COLSPAN=3 HEIGHT=12></TD> <TD ROWSPAN=5></TD></TR><TR> <TD COLSPAN=3></TD></TR><TR><TD></TD> <TD><FONT FACE='Times New Roman, Times, Serif' SIZE=3>Using an SOM to discover clusters in your data can often lead to a problem in trying to extrapolate a meaning for each of the groupings. There are two ways of dealing with this problem. The first involves obtaining the average value for each of the features in the data. For example, if the number of children or customer age is one of the input features, then what you need to obtain is the average value for each of these features in each of the clusters the SOM discovered. The other method involves passing each of the clusters through a machine-learning algorithm in order to generate a set of rules describing the features of that grouping. This second option requires the ability to link the output from an SOM to the inputs of a machine-learning algorithm tool like C4.5, CHAID, or CART, which is possible with some of the new data mining suites, or toolboxes like Clementine.</FONT><FONT FACE='Times New Roman, Times, Serif' SIZE=3 COLOR=#FFFF00><!-- break --></FONT></TD><TD></TD></TR><TR> <TD COLSPAN=3></TD></TR><TR> <TD COLSPAN=3 HEIGHT=1></TD></TR></TABLE><A NAME='{353}'/></FORM></P></TD></TR></TABLE><P><FONT SIZE=0 COLOR=WHITE></CENTER><A NAME="bottom"> </A><!-- netLibrary.com Copyright Notice --> </BODY></HTML>⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -