Mercurial > hg > camir-aes2014

comparison toolboxes/FullBNT-1.0.7/nethelp3.3/netopt.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 netopt
5 </title>
6 </head>
7 <body>
8 <H1> netopt
9 </H1>
10 <h2>
11 Purpose
12 </h2>
13 Optimize the weights in a network model.
14
15 <p><h2>
16 Synopsis
17 </h2>
18 <PRE>
19 [net, options] = netopt(net, options, x, t, alg)
20 [net, options, varargout] = netopt(net, options, x, t, alg)
21 </PRE>
22
23
24 <p><h2>
25 Description
26 </h2>
27
28 <p><CODE>netopt</CODE> is a helper function which facilitates the training of
29 networks using the general purpose optimizers as well as sampling from the
30 posterior distribution of parameters using general purpose Markov chain
31 Monte Carlo sampling algorithms. It can be used with any function that
32 searches in parameter space using error and gradient functions.
33
34 <p><CODE>[net, options] = netopt(net, options, x, t, alg)</CODE> takes a network
35 data structure <CODE>net</CODE>, together with a vector <CODE>options</CODE> of
36 parameters governing the behaviour of the optimization algorithm, a
37 matrix <CODE>x</CODE> of input vectors and a matrix <CODE>t</CODE> of target
38 vectors, and returns the trained network as well as an updated
39 <CODE>options</CODE> vector. The string <CODE>alg</CODE> determines which optimization
40 algorithm (<CODE>conjgrad</CODE>, <CODE>quasinew</CODE>, <CODE>scg</CODE>, etc.) or Monte
41 Carlo algorithm (such as <CODE>hmc</CODE>) will be used.
42
43 <p><CODE>[net, options, varargout] = netopt(net, options, x, t, alg)</CODE>
44 also returns any additional return values from the optimisation algorithm.
45
46 <p><h2>
47 Examples
48 </h2>
49 Suppose we create a 4-input, 3 hidden unit, 2-output feed-forward
50 network using <CODE>net = mlp(4, 3, 2, 'linear')</CODE>. We can then train
51 the network with the scaled conjugate gradient algorithm by using
52 <CODE>net = netopt(net, options, x, t, 'scg')</CODE> where <CODE>x</CODE> and
53 <CODE>t</CODE> are the input and target data matrices respectively, and the
54 options vector is set appropriately for <CODE>scg</CODE>.
55
56 <p>If we also wish to plot the learning curve, we can use the additional
57 return value <CODE>errlog</CODE> given by <CODE>scg</CODE>:
58 <PRE>
59
60 [net, options, errlog] = netopt(net, options, x, t, 'scg');
61 </PRE>
62
63
64 <p><h2>
65 See Also
66 </h2>
67 <CODE><a href="netgrad.htm">netgrad</a></CODE>, <CODE><a href="bfgs.htm">bfgs</a></CODE>, <CODE><a href="conjgrad.htm">conjgrad</a></CODE>, <CODE><a href="graddesc.htm">graddesc</a></CODE>, <CODE><a href="hmc.htm">hmc</a></CODE>, <CODE><a href="scg.htm">scg</a></CODE><hr>
68 <b>Pages:</b>
69 <a href="index.htm">Index</a>
70 <hr>
71 <p>Copyright (c) Ian T Nabney (1996-9)
72
73
74 </body>
75 </html>