final_pred.m

gaussian procession 原程序,用于解决分类问题,有兴趣可以看看啊

function [mean_pred] = final_pred(filename, reject, gsmp, parvec)

% calculate the final predictions of the mean of the softmax
% 
% [mean_pred] = final_pred(filename, reject, gsmp, parvec)
%
% filename : prefix of files that store the means and covariances
% reject   : number of initial accepted samples to be rejected
% gsmp     : number of samples in the Gaussian Softmax average
% parvec   : description vector

%            Matlab code for Gaussian Processes for Classification:
%                      GPCLASS version 0.2  10 Nov 97
%       Copyright (c) David Barber and Christopher K I Williams (1997)

%    This program is free software; you can redistribute it and/or modify
%    it under the terms of the GNU General Public License as published by
%    the Free Software Foundation; either version 2 of the License, or
%    any later version.
%
%    This program is distributed in the hope that it will be useful,
%    but WITHOUT ANY WARRANTY; without even the implied warranty of
%    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
%    GNU General Public License for more details.
%
%    You should have received a copy of the GNU General Public License
%    along with this program; if not, write to the Free Software
%    Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.

[m ntr nte ntrte] = unpak(parvec);

fidm= fopen([filename '.eta'], 'r');  % stored means
fidv = fopen([filename,'.var'], 'r'); % stored covariance matrices

% Calculate the average outputs over the samples

av = zeros(ntrte,m);
means = fscanf(fidm,'%f', [m*ntrte,inf]);    % matrix of the GP eta predictions
t = size(means,2);			% number of theta values.

for i=1+reject:t
  % loop over the MCMC points
  % get the mean of the gaussian:
  mvecs = deaugyout(means(:,i),m,ntr,nte);	% de-augment the vector

  for n = 1:ntrte;				% loop over the datapoints
%    fprintf(1,'Hyperparameter sample = %d, datapoint = %d\n',i,n)
    % get the mean and covarince matrix for each datapoint.  
    mvec = mvecs(n,:);
    covmat = fscanf(fidv,'%f',[m,m]);
    av(n,:) = av(n,:) + msigint(mvec',covmat,gsmp)';
  end
end

mean_pred = av./(t-reject);		% prediction after mean pi space

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

function s  = msigint(m,C,gsmp)

% m is the mean vector, C the covariance matrix. This routine calculates the
% integral over a gaussian of a softmax function by sampling methods.

av = zeros(size(m,1),size(m,2));
for j = 1:gsmp
  y = mdgauss_smp(m,C);			% get a sample from the gaussian
  av = av + softmax(y);
end

s = av./gsmp;

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

function s  = softmax(y,i)

% returns the vector of softmax values, with components y(i)/sum(y(j))

s = exp(y)./sum(exp(y));

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

function d = deaugyout(vec,m,ntr,nte)

% global # classes and # training and test points

% augmented yeta vectors are stored in the form
% class1:train_points, class2:train_points, class1:test, class2:test ....
% want to return this in the form

% class1:train, class2:train
% class1:test , class2:test

d = zeros(ntr+nte,m);

trainbit = vec(1:m*ntr);
testbit = vec(m*ntr+1:length(vec));

d(1:ntr,:) = reshape(trainbit,ntr,m);
d(ntr+1:ntr+nte,:) = reshape(testbit,nte,m);

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
function s=mdgauss_smp(mean,S)

% S is the covariance matrix

A = chol(S)';

n=length(mean);

if size(mean,2)>size(mean,1)
  mean = mean';
end

z=randn(n,1);

s=mean+A*z;