Mercurial > hg > camir-aes2014

diff toolboxes/FullBNT-1.0.7/nethelp3.3/glmhess.htm @ 0:e9a9cd732c1e tip

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
first hg version after svn
author wolffd
date Tue, 10 Feb 2015 15:05:51 +0000
parents
children
line wrap: on
line diff
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/toolboxes/FullBNT-1.0.7/nethelp3.3/glmhess.htm	Tue Feb 10 15:05:51 2015 +0000
@@ -0,0 +1,74 @@
+<html>
+<head>
+<title>
+Netlab Reference Manual glmhess
+</title>
+</head>
+<body>
+<H1> glmhess
+</H1>
+<h2>
+Purpose
+</h2>
+Evaluate the Hessian matrix for a generalised linear model.
+
+<p><h2>
+Synopsis
+</h2>
+<PRE>
+h = glmhess(net, x, t)
+[h, hdata] = glmhess(net, x, t)
+h = glmhess(net, x, t, hdata)
+</PRE>
+
+
+<p><h2>
+Description
+</h2>
+<CODE>h = glmhess(net, x, t)</CODE> takes a GLM network data structure <CODE>net</CODE>,  
+a matrix <CODE>x</CODE> of input values, and a matrix <CODE>t</CODE> of target
+values and returns the full Hessian matrix <CODE>h</CODE> corresponding to
+the second derivatives of the negative log posterior distribution,
+evaluated for the current weight and bias values as defined by
+<CODE>net</CODE>. Note that the target data is not required in the calculation,
+but is included to make the interface uniform with <CODE>nethess</CODE>.  For
+linear and logistic outputs, the computation is very simple and is 
+done (in effect) in one line in <CODE>glmtrain</CODE>.
+
+<p><CODE>[h, hdata] = glmhess(net, x, t)</CODE> returns both the Hessian matrix
+<CODE>h</CODE> and the contribution <CODE>hdata</CODE> arising from the data dependent
+term in the Hessian.
+
+<p><CODE>h = glmhess(net, x, t, hdata)</CODE> takes a network data structure
+<CODE>net</CODE>, a matrix <CODE>x</CODE> of input values, and a matrix <CODE>t</CODE> of 
+target values, together with the contribution <CODE>hdata</CODE> arising from
+the data dependent term in the Hessian, and returns the full Hessian
+matrix <CODE>h</CODE> corresponding to the second derivatives of the negative
+log posterior distribution. This version saves computation time if
+<CODE>hdata</CODE> has already been evaluated for the current weight and bias
+values.
+
+<p><h2>
+Example
+</h2>
+The Hessian matrix is used by <CODE>glmtrain</CODE> to take a Newton step for
+softmax outputs.
+<PRE>
+
+Hessian = glmhess(net, x, t);
+deltaw = -gradient*pinv(Hessian);
+</PRE>
+
+
+<p><h2>
+See Also
+</h2>
+<CODE><a href="glm.htm">glm</a></CODE>, <CODE><a href="glmtrain.htm">glmtrain</a></CODE>, <CODE><a href="hesschek.htm">hesschek</a></CODE>, <CODE><a href="nethess.htm">nethess</a></CODE><hr>
+<b>Pages:</b>
+<a href="index.htm">Index</a>
+<hr>
+<p>Copyright (c) Ian T Nabney (1996-9)
+
+
+</body>
+</html>
\ No newline at end of file