Mercurial > hg > camir-aes2014

comparison toolboxes/FullBNT-1.0.7/nethelp3.3/glmtrain.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
comparison
equal deleted inserted replaced
-1:000000000000 0:e9a9cd732c1e
1 <html>
2 <head>
3 <title>
4 Netlab Reference Manual glmtrain
5 </title>
6 </head>
7 <body>
8 <H1> glmtrain
9 </H1>
10 <h2>
11 Purpose
12 </h2>
13 Specialised training of generalized linear model
14
15 <p><h2>
16 Description
17 </h2>
18 <CODE>net = glmtrain(net, options, x, t)</CODE> uses
19 the iterative reweighted least squares (IRLS)
20 algorithm to set the weights in the generalized linear model structure
21 <CODE>net</CODE>. This is a more efficient alternative to using <CODE>glmerr</CODE>
22 and <CODE>glmgrad</CODE> and a non-linear optimisation routine through
23 <CODE>netopt</CODE>.
24 Note that for linear outputs, a single pass through the
25 algorithm is all that is required, since the error function is quadratic in
26 the weights. The algorithm also handles scalar <CODE>alpha</CODE> and <CODE>beta</CODE>
27 terms. If you want to use more complicated priors, you should use
28 general-purpose non-linear optimisation algorithms.
29
30 <p>For logistic and softmax outputs, general priors can be handled, although
31 this requires the pseudo-inverse of the Hessian, giving up the better
32 conditioning and some of the speed advantage of the normal form equations.
33
34 <p>The error function value at the final set of weights is returned
35 in <CODE>options(8)</CODE>.
36 Each row of <CODE>x</CODE> corresponds to one
37 input vector and each row of <CODE>t</CODE> corresponds to one target vector.
38
39 <p>The optional parameters have the following interpretations.
40
41 <p><CODE>options(1)</CODE> is set to 1 to display error values during training.
42 If <CODE>options(1)</CODE> is set to 0,
43 then only warning messages are displayed. If <CODE>options(1)</CODE> is -1,
44 then nothing is displayed.
45
46 <p><CODE>options(2)</CODE> is a measure of the precision required for the value
47 of the weights <CODE>w</CODE> at the solution.
48
49 <p><CODE>options(3)</CODE> is a measure of the precision required of the objective
50 function at the solution. Both this and the previous condition must be
51 satisfied for termination.
52
53 <p><CODE>options(5)</CODE> is set to 1 if an approximation to the Hessian (which assumes
54 that all outputs are independent) is used for softmax outputs. With the default
55 value of 0 the exact Hessian (which is more expensive to compute) is used.
56
57 <p><CODE>options(14)</CODE> is the maximum number of iterations for the IRLS algorithm;
58 default 100.
59
60 <p><h2>
61 See Also
62 </h2>
63 <CODE><a href="glm.htm">glm</a></CODE>, <CODE><a href="glmerr.htm">glmerr</a></CODE>, <CODE><a href="glmgrad.htm">glmgrad</a></CODE><hr>
64 <b>Pages:</b>
65 <a href="index.htm">Index</a>
66 <hr>
67 <p>Copyright (c) Ian T Nabney (1996-9)
68
69
70 </body>
71 </html>