Mercurial > hg > camir-aes2014
diff toolboxes/FullBNT-1.0.7/netlab3.3/rbfhess.m @ 0:e9a9cd732c1e tip
first hg version after svn
author | wolffd |
---|---|
date | Tue, 10 Feb 2015 15:05:51 +0000 |
parents | |
children |
line wrap: on
line diff
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/toolboxes/FullBNT-1.0.7/netlab3.3/rbfhess.m Tue Feb 10 15:05:51 2015 +0000 @@ -0,0 +1,91 @@ +function [h, hdata] = rbfhess(net, x, t, hdata) +%RBFHESS Evaluate the Hessian matrix for RBF network. +% +% Description +% H = RBFHESS(NET, X, T) takes an RBF network data structure NET, a +% matrix X of input values, and a matrix T of target values and returns +% the full Hessian matrix H corresponding to the second derivatives of +% the negative log posterior distribution, evaluated for the current +% weight and bias values as defined by NET. Currently, the +% implementation only computes the Hessian for the output layer +% weights. +% +% [H, HDATA] = RBFHESS(NET, X, T) returns both the Hessian matrix H and +% the contribution HDATA arising from the data dependent term in the +% Hessian. +% +% H = RBFHESS(NET, X, T, HDATA) takes a network data structure NET, a +% matrix X of input values, and a matrix T of target values, together +% with the contribution HDATA arising from the data dependent term in +% the Hessian, and returns the full Hessian matrix H corresponding to +% the second derivatives of the negative log posterior distribution. +% This version saves computation time if HDATA has already been +% evaluated for the current weight and bias values. +% +% See also +% MLPHESS, HESSCHEK, EVIDENCE +% + +% Copyright (c) Ian T Nabney (1996-2001) + +% Check arguments for consistency +errstring = consist(net, 'rbf', x, t); +if ~isempty(errstring); + error(errstring); +end + +if nargin == 3 + % Data term in Hessian needs to be computed + [a, z] = rbffwd(net, x); + hdata = datahess(net, z, t); +end + +% Add in effect of regularisation +[h, hdata] = hbayes(net, hdata); + +% Sub-function to compute data part of Hessian +function hdata = datahess(net, z, t) + +% Only works for output layer Hessian currently +if (isfield(net, 'mask') & ~any(net.mask(... + 1:(net.nwts - net.nout*(net.nhidden+1))))) + hdata = zeros(net.nwts); + ndata = size(z, 1); + out_hess = [z ones(ndata, 1)]'*[z ones(ndata, 1)]; + for j = 1:net.nout + hdata = rearrange_hess(net, j, out_hess, hdata); + end +else + error('Output layer Hessian only.'); +end +return + +% Sub-function to rearrange Hessian matrix +function hdata = rearrange_hess(net, j, out_hess, hdata) + +% Because all the biases come after all the input weights, +% we have to rearrange the blocks that make up the network Hessian. +% This function assumes that we are on the jth output and that all outputs +% are independent. + +% Start of bias weights block +bb_start = net.nwts - net.nout + 1; +% Start of weight block for jth output +ob_start = net.nwts - net.nout*(net.nhidden+1) + (j-1)*net.nhidden... + + 1; +% End of weight block for jth output +ob_end = ob_start + net.nhidden - 1; +% Index of bias weight +b_index = bb_start+(j-1); +% Put input weight block in right place +hdata(ob_start:ob_end, ob_start:ob_end) = out_hess(1:net.nhidden, ... + 1:net.nhidden); +% Put second derivative of bias weight in right place +hdata(b_index, b_index) = out_hess(net.nhidden+1, net.nhidden+1); +% Put cross terms (input weight v bias weight) in right place +hdata(b_index, ob_start:ob_end) = out_hess(net.nhidden+1, ... + 1:net.nhidden); +hdata(ob_start:ob_end, b_index) = out_hess(1:net.nhidden, ... + net.nhidden+1); + +return \ No newline at end of file