|
wolffd@0
|
1 <html>
|
|
wolffd@0
|
2 <head>
|
|
wolffd@0
|
3 <title>
|
|
wolffd@0
|
4 Netlab Reference Manual rbftrain
|
|
wolffd@0
|
5 </title>
|
|
wolffd@0
|
6 </head>
|
|
wolffd@0
|
7 <body>
|
|
wolffd@0
|
8 <H1> rbftrain
|
|
wolffd@0
|
9 </H1>
|
|
wolffd@0
|
10 <h2>
|
|
wolffd@0
|
11 Purpose
|
|
wolffd@0
|
12 </h2>
|
|
wolffd@0
|
13 Two stage training of RBF network.
|
|
wolffd@0
|
14
|
|
wolffd@0
|
15 <p><h2>
|
|
wolffd@0
|
16 Description
|
|
wolffd@0
|
17 </h2>
|
|
wolffd@0
|
18 <CODE>net = rbftrain(net, options, x, t)</CODE> uses a
|
|
wolffd@0
|
19 two stage training
|
|
wolffd@0
|
20 algorithm to set the weights in the RBF model structure <CODE>net</CODE>.
|
|
wolffd@0
|
21 Each row of <CODE>x</CODE> corresponds to one
|
|
wolffd@0
|
22 input vector and each row of <CODE>t</CODE> contains the corresponding target vector.
|
|
wolffd@0
|
23 The centres are determined by fitting a Gaussian mixture model
|
|
wolffd@0
|
24 with circular covariances using the EM algorithm through a call to
|
|
wolffd@0
|
25 <CODE>rbfsetbf</CODE>. (The mixture model is
|
|
wolffd@0
|
26 initialised using a small number of iterations of the K-means algorithm.)
|
|
wolffd@0
|
27 If the activation functions are Gaussians, then the basis function widths
|
|
wolffd@0
|
28 are then set to the maximum inter-centre squared distance.
|
|
wolffd@0
|
29
|
|
wolffd@0
|
30 <p>For linear outputs,
|
|
wolffd@0
|
31 the hidden to output
|
|
wolffd@0
|
32 weights that give rise to the least squares solution
|
|
wolffd@0
|
33 can then be determined using the pseudo-inverse. For neuroscale outputs,
|
|
wolffd@0
|
34 the hidden to output weights are determined using the iterative shadow
|
|
wolffd@0
|
35 targets algorithm.
|
|
wolffd@0
|
36 Although this two stage
|
|
wolffd@0
|
37 procedure may not give solutions with as low an error as using general
|
|
wolffd@0
|
38 purpose non-linear optimisers, it is much faster.
|
|
wolffd@0
|
39
|
|
wolffd@0
|
40 <p>The options vector may have two rows: if this is the case, then the second row
|
|
wolffd@0
|
41 is passed to <CODE>rbfsetbf</CODE>, which allows the user to specify a different
|
|
wolffd@0
|
42 number iterations for RBF and GMM training.
|
|
wolffd@0
|
43 The optional parameters to <CODE>rbftrain</CODE> have the following interpretations.
|
|
wolffd@0
|
44
|
|
wolffd@0
|
45 <p><CODE>options(1)</CODE> is set to 1 to display error values during EM training.
|
|
wolffd@0
|
46
|
|
wolffd@0
|
47 <p><CODE>options(2)</CODE> is a measure of the precision required for the value
|
|
wolffd@0
|
48 of the weights <CODE>w</CODE> at the solution.
|
|
wolffd@0
|
49
|
|
wolffd@0
|
50 <p><CODE>options(3)</CODE> is a measure of the precision required of the objective
|
|
wolffd@0
|
51 function at the solution. Both this and the previous condition must be
|
|
wolffd@0
|
52 satisfied for termination.
|
|
wolffd@0
|
53
|
|
wolffd@0
|
54 <p><CODE>options(5)</CODE> is set to 1 if the basis functions parameters should remain
|
|
wolffd@0
|
55 unchanged; default 0.
|
|
wolffd@0
|
56
|
|
wolffd@0
|
57 <p><CODE>options(6)</CODE> is set to 1 if the output layer weights should be should
|
|
wolffd@0
|
58 set using PCA. This is only relevant for Neuroscale outputs; default 0.
|
|
wolffd@0
|
59
|
|
wolffd@0
|
60 <p><CODE>options(14)</CODE> is the maximum number of iterations for the shadow
|
|
wolffd@0
|
61 targets algorithm;
|
|
wolffd@0
|
62 default 100.
|
|
wolffd@0
|
63
|
|
wolffd@0
|
64 <p><h2>
|
|
wolffd@0
|
65 Example
|
|
wolffd@0
|
66 </h2>
|
|
wolffd@0
|
67 The following example creates an RBF network and then trains it:
|
|
wolffd@0
|
68 <PRE>
|
|
wolffd@0
|
69
|
|
wolffd@0
|
70 net = rbf(1, 4, 1, 'gaussian');
|
|
wolffd@0
|
71 options(1, :) = foptions;
|
|
wolffd@0
|
72 options(2, :) = foptions;
|
|
wolffd@0
|
73 options(2, 14) = 10; % 10 iterations of EM
|
|
wolffd@0
|
74 options(2, 5) = 1; % Check for covariance collapse in EM
|
|
wolffd@0
|
75 net = rbftrain(net, options, x, t);
|
|
wolffd@0
|
76 </PRE>
|
|
wolffd@0
|
77
|
|
wolffd@0
|
78
|
|
wolffd@0
|
79 <p><h2>
|
|
wolffd@0
|
80 See Also
|
|
wolffd@0
|
81 </h2>
|
|
wolffd@0
|
82 <CODE><a href="rbf.htm">rbf</a></CODE>, <CODE><a href="rbferr.htm">rbferr</a></CODE>, <CODE><a href="rbffwd.htm">rbffwd</a></CODE>, <CODE><a href="rbfgrad.htm">rbfgrad</a></CODE>, <CODE><a href="rbfpak.htm">rbfpak</a></CODE>, <CODE><a href="rbfunpak.htm">rbfunpak</a></CODE>, <CODE><a href="rbfsetbf.htm">rbfsetbf</a></CODE><hr>
|
|
wolffd@0
|
83 <b>Pages:</b>
|
|
wolffd@0
|
84 <a href="index.htm">Index</a>
|
|
wolffd@0
|
85 <hr>
|
|
wolffd@0
|
86 <p>Copyright (c) Ian T Nabney (1996-9)
|
|
wolffd@0
|
87
|
|
wolffd@0
|
88
|
|
wolffd@0
|
89 </body>
|
|
wolffd@0
|
90 </html> |
