wolffd@0
|
1 function [net, gamma, logev] = evidence_weighted(net, x, t, eso_w, num)
|
wolffd@0
|
2 %EVIDENCE Re-estimate hyperparameters using evidence approximation.
|
wolffd@0
|
3 %
|
wolffd@0
|
4 % Description
|
wolffd@0
|
5 % [NET] = EVIDENCE(NET, X, T) re-estimates the hyperparameters ALPHA
|
wolffd@0
|
6 % and BETA by applying Bayesian re-estimation formulae for NUM
|
wolffd@0
|
7 % iterations. The hyperparameter ALPHA can be a simple scalar
|
wolffd@0
|
8 % associated with an isotropic prior on the weights, or can be a vector
|
wolffd@0
|
9 % in which each component is associated with a group of weights as
|
wolffd@0
|
10 % defined by the INDEX matrix in the NET data structure. These more
|
wolffd@0
|
11 % complex priors can be set up for an MLP using MLPPRIOR. Initial
|
wolffd@0
|
12 % values for the iterative re-estimation are taken from the network
|
wolffd@0
|
13 % data structure NET passed as an input argument, while the return
|
wolffd@0
|
14 % argument NET contains the re-estimated values.
|
wolffd@0
|
15 %
|
wolffd@0
|
16 % [NET, GAMMA, LOGEV] = EVIDENCE(NET, X, T, NUM) allows the re-
|
wolffd@0
|
17 % estimation formula to be applied for NUM cycles in which the re-
|
wolffd@0
|
18 % estimated values for the hyperparameters from each cycle are used to
|
wolffd@0
|
19 % re-evaluate the Hessian matrix for the next cycle. The return value
|
wolffd@0
|
20 % GAMMA is the number of well-determined parameters and LOGEV is the
|
wolffd@0
|
21 % log of the evidence.
|
wolffd@0
|
22 %
|
wolffd@0
|
23 % See also
|
wolffd@0
|
24 % MLPPRIOR, NETGRAD, NETHESS, DEMEV1, DEMARD
|
wolffd@0
|
25 %
|
wolffd@0
|
26
|
wolffd@0
|
27 % Copyright (c) Ian T Nabney (1996-9)
|
wolffd@0
|
28
|
wolffd@0
|
29 errstring = consist(net, '', x, t);
|
wolffd@0
|
30 if ~isempty(errstring)
|
wolffd@0
|
31 error(errstring);
|
wolffd@0
|
32 end
|
wolffd@0
|
33
|
wolffd@0
|
34 ndata = size(x, 1);
|
wolffd@0
|
35 if nargin == 4
|
wolffd@0
|
36 num = 1;
|
wolffd@0
|
37 end
|
wolffd@0
|
38
|
wolffd@0
|
39 if isfield(net,'beta')
|
wolffd@0
|
40 beta = net.beta;
|
wolffd@0
|
41 else
|
wolffd@0
|
42 beta = 1;
|
wolffd@0
|
43 end;
|
wolffd@0
|
44
|
wolffd@0
|
45 % Extract weights from network
|
wolffd@0
|
46 pakstr = [net.type, 'pak'];
|
wolffd@0
|
47 w = feval(pakstr, net);
|
wolffd@0
|
48
|
wolffd@0
|
49 % Evaluate data-dependent contribution to the Hessian matrix.
|
wolffd@0
|
50 [h, dh] = nethess_weighted(w, net, x, t, eso_w);
|
wolffd@0
|
51
|
wolffd@0
|
52 % Now set the negative eigenvalues to zero.
|
wolffd@0
|
53 [evec, evl] = eig(dh);
|
wolffd@0
|
54 evl = evl.*(evl > 0);
|
wolffd@0
|
55 % safe_evl is used to avoid taking log of zero
|
wolffd@0
|
56 safe_evl = evl + eps.*(evl <= 0);
|
wolffd@0
|
57
|
wolffd@0
|
58 % Do the re-estimation.
|
wolffd@0
|
59 for k = 1 : num
|
wolffd@0
|
60 [e, edata, eprior] = neterr_weighted(w, net, x, t, eso_w);
|
wolffd@0
|
61 h = nethess_weighted(w, net, x, t, eso_w, dh);
|
wolffd@0
|
62 % Re-estimate alpha.
|
wolffd@0
|
63 if size(net.alpha) == [1 1]
|
wolffd@0
|
64 % Evaluate number of well-determined parameters.
|
wolffd@0
|
65 if k == 1
|
wolffd@0
|
66 % Form vector of eigenvalues
|
wolffd@0
|
67 evl = diag(evl);
|
wolffd@0
|
68 safe_evl = diag(safe_evl);
|
wolffd@0
|
69 end
|
wolffd@0
|
70 B = beta*evl;
|
wolffd@0
|
71 gamma = sum(B./(B + net.alpha));
|
wolffd@0
|
72 net.alpha = 0.5*gamma/eprior;
|
wolffd@0
|
73
|
wolffd@0
|
74 % Partially evaluate log evidence
|
wolffd@0
|
75 logev = e - 0.5*sum(log(safe_evl)) + 0.5*net.nwts*log(net.alpha) - ...
|
wolffd@0
|
76 0.5*ndata*log(2*pi);
|
wolffd@0
|
77 else
|
wolffd@0
|
78 ngroups = size(net.alpha, 1);
|
wolffd@0
|
79 gams = zeros(1, ngroups);
|
wolffd@0
|
80 logas = zeros(1, ngroups);
|
wolffd@0
|
81 traces = zeros(1, ngroups);
|
wolffd@0
|
82 % Reconstruct data hessian with negative eigenvalues set to zero.
|
wolffd@0
|
83 dh = evec*evl*evec';
|
wolffd@0
|
84 hinv = inv(nethess_weighted(w, net, x, t, eso_w, dh));
|
wolffd@0
|
85 for m = 1 : ngroups
|
wolffd@0
|
86 group_nweights = sum(net.index(:, m));
|
wolffd@0
|
87 gams(m) = group_nweights - ...
|
wolffd@0
|
88 net.alpha(m)*sum(diag(hinv).*net.index(:,m));
|
wolffd@0
|
89 net.alpha(m) = real(gams(m)/(2*eprior(m)));
|
wolffd@0
|
90 % Weight alphas by number of weights in group
|
wolffd@0
|
91 logas(m) = 0.5*group_nweights*log(net.alpha(m));
|
wolffd@0
|
92 % Compute sum of evalues corresponding to group
|
wolffd@0
|
93 traces(m) = sum(log(safe_evl*net.index(:,m)));
|
wolffd@0
|
94 end
|
wolffd@0
|
95 gamma = sum(gams, 2);
|
wolffd@0
|
96 logev = e - 0.5*sum(traces) + sum(logas) - 0.5*ndata*log(2*pi);
|
wolffd@0
|
97 end
|
wolffd@0
|
98 % Re-estimate beta.
|
wolffd@0
|
99 if isfield(net, 'beta')
|
wolffd@0
|
100 net.beta = 0.5*(net.nout*ndata - gamma)/edata;
|
wolffd@0
|
101 end
|
wolffd@0
|
102 logev = logev + 0.5*ndata*log(beta);
|
wolffd@0
|
103 end
|
wolffd@0
|
104
|