📄 如何训练.html
字号:
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN">
<HTML xmlns:o = "urn:schemas-microsoft-com:office:office" xmlns:v =
"urn:schemas-microsoft-com:vml"><HEAD><TITLE></TITLE>
<META http-equiv=Content-Type content="text/html; charset=GB2312">
<META content="MSHTML 6.00.2900.3157" name=GENERATOR></HEAD>
<BODY>
<P class=MsoNormal
style="MARGIN: 0cm 0cm 0pt; TEXT-INDENT: 24pt; TEXT-ALIGN: left; mso-char-indent-count: 2.0; mso-layout-grid-align: none"
align=left><SPAN
style="FONT-SIZE: 12pt; FONT-FAMILY: 宋体; mso-bidi-font-family: 黑体; mso-font-kerning: 0pt">给定一个输入序列<SPAN lang=EN-US><STRONG>X</STRONG></SPAN>，条件随机域</SPAN><SPAN lang=EN-US
style="FONT-SIZE: 12pt; mso-bidi-font-size: 9.0pt; mso-font-kerning: 0pt">(CRF)</SPAN><SPAN
style="FONT-SIZE: 12pt; FONT-FAMILY: 宋体; mso-bidi-font-family: 黑体; mso-font-kerning: 0pt">模型的目标是找到使如下条件概率最大化的状态序列<IMG
src="file://C:\Documents and Settings\Administrator\My Documents\CRF帮助文档\train2.JPG"
align=baseline></SPAN><SPAN
style="FONT-FAMILY: 宋体; mso-ascii-font-family: 'Times New Roman'; mso-hansi-font-family: 'Times New Roman'">：<BR>
</SPAN><SPAN lang=EN-US><SPAN style="mso-spacerun: yes"><IMG
src="file://C:\Documents and Settings\Administrator\My Documents\CRF帮助文档\train1.JPG"
align=baseline> </SPAN><SPAN
style="mso-tab-count: 2"> </SPAN><SPAN
style="mso-spacerun: yes"> </SPAN><SPAN
style="mso-tab-count: 2"> <BR></SPAN></SPAN><SPAN lang=EN-US
style="FONT-SIZE: 12pt; FONT-FAMILY: 宋体; mso-bidi-font-family: 黑体; mso-font-kerning: 0pt"><SPAN style="mso-spacerun: yes"> </SPAN></SPAN><SPAN lang=EN-US
style="FONT-SIZE: 12pt; mso-bidi-font-size: 9.0pt; mso-font-kerning: 0pt"><STRONG>Viterbi</STRONG></SPAN><SPAN
style="FONT-SIZE: 12pt; FONT-FAMILY: 宋体; mso-bidi-font-family: 黑体; mso-font-kerning: 0pt">算法通常用来解决此类问题，若需要获取</SPAN><SPAN lang=EN-US
style="FONT-SIZE: 12pt; mso-bidi-font-size: 9.0pt; mso-font-kerning: 0pt"><STRONG>N-best</STRONG></SPAN><SPAN
style="FONT-SIZE: 12pt; FONT-FAMILY: 宋体; mso-bidi-font-family: 黑体; mso-font-kerning: 0pt">序列，则需要一种修改后的</SPAN><SPAN lang=EN-US
style="FONT-SIZE: 12pt; mso-bidi-font-size: 9.0pt; mso-font-kerning: 0pt"><STRONG>Viterbi</STRONG></SPAN><SPAN
style="FONT-SIZE: 12pt; FONT-FAMILY: 宋体; mso-bidi-font-family: 黑体; mso-font-kerning: 0pt">算法。<SPAN lang=EN-US><o:p></o:p></SPAN></SPAN></P>
<P class=MsoNormal
style="MARGIN: 0cm 0cm 0pt; TEXT-INDENT: 24pt; TEXT-ALIGN: left; mso-char-indent-count: 2.0; mso-layout-grid-align: none"
align=left><SPAN
style="FONT-SIZE: 12pt; FONT-FAMILY: 宋体; mso-bidi-font-family: 黑体; mso-font-kerning: 0pt">为了能在给定输入序列的情况下估计模型参数</SPAN><SPAN lang=EN-US><SPAN
style="POSITION: relative; TOP: 5pt; mso-text-raise: -5.0pt"><IMG
src="file://C:\Documents and Settings\Administrator\My Documents\CRF帮助文档\Lameda.JPG"
align=baseline></SPAN></SPAN><SPAN
style="FONT-FAMILY: 宋体; mso-ascii-font-family: 'Times New Roman'; mso-hansi-font-family: 'Times New Roman'">，</SPAN><SPAN
style="FONT-SIZE: 12pt; FONT-FAMILY: 宋体; mso-bidi-font-family: 黑体; mso-font-kerning: 0pt">条件随机域模型的最大后验概率训练可以用最大似然的方法把事先准备好的训练语料<IMG
src="file://C:\Documents and Settings\Administrator\My Documents\CRF帮助文档\train3.JPG"
align=baseline></SPAN><SPAN
style="FONT-SIZE: 12pt; FONT-FAMILY: 宋体; mso-bidi-font-family: 黑体; mso-font-kerning: 0pt">构成的集合的条件概率最大化：<BR>
</SPAN><SPAN lang=EN-US><SPAN style="mso-tab-count: 5"><IMG
src="file://C:\Documents and Settings\Administrator\My Documents\CRF帮助文档\train4.JPG"
align=baseline><BR></SPAN></SPAN><SPAN
style="FONT-SIZE: 12pt; FONT-FAMILY: 宋体; mso-bidi-font-family: 黑体; mso-font-kerning: 0pt">为了避免过拟合，在条件随机域的公式</SPAN><SPAN
style="FONT-SIZE: 12pt; FONT-FAMILY: 宋体; mso-bidi-font-family: 黑体; mso-font-kerning: 0pt">中需要加入一个惩罚因子，代入得到如下公式：<BR><BR>
<IMG
src="file://C:\Documents and Settings\Administrator\My Documents\CRF帮助文档\train5.JPG"
align=baseline><BR><SPAN
style="FONT-SIZE: 12pt; FONT-FAMILY: 宋体; mso-bidi-font-family: 黑体; mso-ansi-language: EN-US; mso-fareast-language: ZH-CN; mso-bidi-language: AR-SA">对于传统的最大熵模型的训练算法，包括</SPAN><SPAN lang=EN-US
style="FONT-SIZE: 12pt; FONT-FAMILY: 'Times New Roman'; mso-bidi-font-size: 9.0pt; mso-ansi-language: EN-US; mso-fareast-language: ZH-CN; mso-bidi-language: AR-SA; mso-fareast-font-family: 宋体">GIS</SPAN><SPAN
style="FONT-SIZE: 12pt; FONT-FAMILY: 宋体; mso-bidi-font-size: 9.0pt; mso-ascii-font-family: 'Times New Roman'; mso-hansi-font-family: 'Times New Roman'; mso-bidi-font-family: 'Times New Roman'; mso-ansi-language: EN-US; mso-fareast-language: ZH-CN; mso-bidi-language: AR-SA">，</SPAN><SPAN lang=EN-US
style="FONT-SIZE: 12pt; FONT-FAMILY: 'Times New Roman'; mso-bidi-font-size: 9.0pt; mso-ansi-language: EN-US; mso-fareast-language: ZH-CN; mso-bidi-language: AR-SA; mso-fareast-font-family: 宋体">IIS</SPAN><SPAN
style="FONT-SIZE: 12pt; FONT-FAMILY: 宋体; mso-bidi-font-family: 黑体; mso-ansi-language: EN-US; mso-fareast-language: ZH-CN; mso-bidi-language: AR-SA">等等<SPAN lang=EN-US>,</SPAN>也能用于条件随机域</SPAN><SPAN lang=EN-US
style="FONT-SIZE: 12pt; FONT-FAMILY: 'Times New Roman'; mso-bidi-font-size: 9.0pt; mso-ansi-language: EN-US; mso-fareast-language: ZH-CN; mso-bidi-language: AR-SA; mso-fareast-font-family: 宋体">(CRF)</SPAN><SPAN
style="FONT-SIZE: 12pt; FONT-FAMILY: 宋体; mso-bidi-font-family: 黑体; mso-ansi-language: EN-US; mso-fareast-language: ZH-CN; mso-bidi-language: AR-SA">的训练。也可以采用爬山</SPAN><SPAN lang=EN-US
style="FONT-SIZE: 12pt; FONT-FAMILY: 'Times New Roman'; mso-bidi-font-size: 9.0pt; mso-ansi-language: EN-US; mso-fareast-language: ZH-CN; mso-bidi-language: AR-SA; mso-fareast-font-family: 宋体">(Hill-climbing)</SPAN><SPAN
style="FONT-SIZE: 12pt; FONT-FAMILY: 宋体; mso-bidi-font-family: 黑体; mso-ansi-language: EN-US; mso-fareast-language: ZH-CN; mso-bidi-language: AR-SA">方法，包括变梯度方法，或二阶的同类方法</SPAN><SPAN lang=EN-US
style="FONT-SIZE: 12pt; FONT-FAMILY: 'Times New Roman'; mso-bidi-font-size: 9.0pt; mso-ansi-language: EN-US; mso-fareast-language: ZH-CN; mso-bidi-language: AR-SA; mso-fareast-font-family: 宋体">L-BFGS</SPAN><SPAN
style="FONT-SIZE: 12pt; FONT-FAMILY: 宋体; mso-bidi-font-family: 黑体; mso-ansi-language: EN-US; mso-fareast-language: ZH-CN; mso-bidi-language: AR-SA">。相对于其他方法，</SPAN><SPAN lang=EN-US
style="FONT-SIZE: 12pt; FONT-FAMILY: 'Times New Roman'; mso-bidi-font-size: 9.0pt; mso-ansi-language: EN-US; mso-fareast-language: ZH-CN; mso-bidi-language: AR-SA; mso-fareast-font-family: 宋体">L-BFGS</SPAN><SPAN
style="FONT-SIZE: 12pt; FONT-FAMILY: 宋体; mso-bidi-font-family: 黑体; mso-ansi-language: EN-US; mso-fareast-language: ZH-CN; mso-bidi-language: AR-SA">具有更快的收敛速度。</SPAN></SPAN></P></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 + -