|
Daniel@0
|
1 <html>
|
|
Daniel@0
|
2 <head>
|
|
Daniel@0
|
3 <title>
|
|
Daniel@0
|
4 Netlab Reference Manual evidence
|
|
Daniel@0
|
5 </title>
|
|
Daniel@0
|
6 </head>
|
|
Daniel@0
|
7 <body>
|
|
Daniel@0
|
8 <H1> evidence
|
|
Daniel@0
|
9 </H1>
|
|
Daniel@0
|
10 <h2>
|
|
Daniel@0
|
11 Purpose
|
|
Daniel@0
|
12 </h2>
|
|
Daniel@0
|
13 Re-estimate hyperparameters using evidence approximation.
|
|
Daniel@0
|
14
|
|
Daniel@0
|
15 <p><h2>
|
|
Daniel@0
|
16 Synopsis
|
|
Daniel@0
|
17 </h2>
|
|
Daniel@0
|
18 <PRE>
|
|
Daniel@0
|
19 [net] = evidence(net, x, t)
|
|
Daniel@0
|
20 [net, gamma, logev] = evidence(net, x, t, num)
|
|
Daniel@0
|
21 </PRE>
|
|
Daniel@0
|
22
|
|
Daniel@0
|
23
|
|
Daniel@0
|
24 <p><h2>
|
|
Daniel@0
|
25 Description
|
|
Daniel@0
|
26 </h2>
|
|
Daniel@0
|
27 <CODE>[net] = evidence(net, x, t)</CODE> re-estimates the
|
|
Daniel@0
|
28 hyperparameters <CODE>alpha</CODE> and <CODE>beta</CODE> by applying Bayesian
|
|
Daniel@0
|
29 re-estimation formulae for <CODE>num</CODE> iterations. The hyperparameter
|
|
Daniel@0
|
30 <CODE>alpha</CODE> can be a simple scalar associated with an isotropic prior
|
|
Daniel@0
|
31 on the weights, or can be a vector in which each component is
|
|
Daniel@0
|
32 associated with a group of weights as defined by the <CODE>index</CODE>
|
|
Daniel@0
|
33 matrix in the <CODE>net</CODE> data structure. These more complex priors can
|
|
Daniel@0
|
34 be set up for an MLP using <CODE>mlpprior</CODE>. Initial values for the iterative
|
|
Daniel@0
|
35 re-estimation are taken from the network data structure <CODE>net</CODE>
|
|
Daniel@0
|
36 passed as an input argument, while the return argument <CODE>net</CODE>
|
|
Daniel@0
|
37 contains the re-estimated values.
|
|
Daniel@0
|
38
|
|
Daniel@0
|
39 <p><CODE>[net, gamma, logev] = evidence(net, x, t, num)</CODE> allows the re-estimation
|
|
Daniel@0
|
40 formula to be applied for <CODE>num</CODE> cycles in which the re-estimated
|
|
Daniel@0
|
41 values for the hyperparameters from each cycle are used to re-evaluate
|
|
Daniel@0
|
42 the Hessian matrix for the next cycle. The return value <CODE>gamma</CODE> is
|
|
Daniel@0
|
43 the number of well-determined parameters and <CODE>logev</CODE> is the log
|
|
Daniel@0
|
44 of the evidence.
|
|
Daniel@0
|
45
|
|
Daniel@0
|
46 <p><h2>
|
|
Daniel@0
|
47 See Also
|
|
Daniel@0
|
48 </h2>
|
|
Daniel@0
|
49 <CODE><a href="mlpprior.htm">mlpprior</a></CODE>, <CODE><a href="netgrad.htm">netgrad</a></CODE>, <CODE><a href="nethess.htm">nethess</a></CODE>, <CODE><a href="demev1.htm">demev1</a></CODE>, <CODE><a href="demard.htm">demard</a></CODE><hr>
|
|
Daniel@0
|
50 <b>Pages:</b>
|
|
Daniel@0
|
51 <a href="index.htm">Index</a>
|
|
Daniel@0
|
52 <hr>
|
|
Daniel@0
|
53 <p>Copyright (c) Ian T Nabney (1996-9)
|
|
Daniel@0
|
54
|
|
Daniel@0
|
55
|
|
Daniel@0
|
56 </body>
|
|
Daniel@0
|
57 </html> |
