[5] | 1 | %WVOTEC Weighted combiner (Adaboost weights)
|
---|
| 2 | %
|
---|
| 3 | % W = WVOTEC(A,V) compute weigths and store
|
---|
| 4 | % W = WVOTEC(V,U) Construct weighted combiner using weights U
|
---|
| 5 | %
|
---|
| 6 | % INPUT
|
---|
| 7 | % A Labeled dataset
|
---|
| 8 | % V Parallel or stacked set of trained classifiers
|
---|
| 9 | % U Set of classifier weights
|
---|
| 10 | %
|
---|
| 11 | % OUTPUT
|
---|
| 12 | % W Combined classifier
|
---|
| 13 | %
|
---|
| 14 | % DESCRIPTION
|
---|
| 15 | % The set of trained classifiers V is combined using weighted
|
---|
| 16 | % majority voting. If given the weights U are used. If not
|
---|
| 17 | % given, the weights are computed from the classification
|
---|
| 18 | % results of the labeled dataset A using 0.5*log((1-E)/E)
|
---|
| 19 | % if E is the classifier error.
|
---|
| 20 | %
|
---|
| 21 | % SEE ALSO
|
---|
| 22 | % MAPPINGS, DATASETS,
|
---|
| 23 |
|
---|
| 24 | % Copyright: R.P.W. Duin, r.p.w.duin@prtools.org
|
---|
| 25 | % Faculty EWI, Delft University of Technology
|
---|
| 26 | % P.O. Box 5031, 2600 GA Delft, The Netherlands
|
---|
| 27 |
|
---|
| 28 | function w = wvotec(a,v)
|
---|
| 29 |
|
---|
| 30 | prtrace(mfilename);
|
---|
| 31 |
|
---|
| 32 | if nargin < 1 | isempty(a)
|
---|
| 33 | w = mapping(mfilename,'untrained');
|
---|
| 34 | elseif nargin < 2
|
---|
| 35 | error('Illegal call')
|
---|
| 36 | elseif isdataset(a) % train or classify
|
---|
| 37 | if ~strcmp(v.mapping_file,mfilename) % training
|
---|
| 38 | if isparallel(v) % parallel combiner
|
---|
| 39 | n = 0;
|
---|
| 40 | e = zeros(1,length(v.data));
|
---|
| 41 | for j=1:length(v.data)
|
---|
| 42 | vv = v.data{j};
|
---|
| 43 | d = a(:,n+1:n+size(vv,1))*vv*classc;
|
---|
| 44 | e(j) = testmc(d);
|
---|
| 45 | n = n+size(vv,1);
|
---|
| 46 | end
|
---|
| 47 | elseif isstacked(v) % stacked combiner
|
---|
| 48 | e = zeros(1,length(v.data));
|
---|
| 49 | for j=1:length(v.data)
|
---|
| 50 | vv = v.data{j};
|
---|
| 51 | e(j) = testmc(a,vv);
|
---|
| 52 | end
|
---|
| 53 | else
|
---|
| 54 | error('Classifier combination expected')
|
---|
| 55 | end
|
---|
| 56 | % Find weights
|
---|
| 57 | L = find(e < 1-max(getprior(a))); % take classifier better than prior
|
---|
| 58 | alf = zeros(1,length(e));
|
---|
| 59 | alf(L) = log((1-e(L))./e(L))/2;
|
---|
| 60 | alf = alf/sum(alf);
|
---|
| 61 | % construct the classifier
|
---|
| 62 | [m,k,c] = getsize(a);
|
---|
| 63 | w = mapping(mfilename,'trained',{v,alf},getlabels(vv),k,c);
|
---|
| 64 | w = setname(w,'Weighted Voting');
|
---|
| 65 | else % testing
|
---|
| 66 | alf = v.data{2}; % get the weights
|
---|
| 67 | u = v.data{1}; % get the set of classifiers
|
---|
| 68 | m = size(a,1);
|
---|
| 69 | dtot = zeros(m,size(v,2));
|
---|
| 70 | if isparallel(u) % parallel combiner
|
---|
| 71 | n = 0;
|
---|
| 72 | for j=1:length(u.data) % weight them
|
---|
| 73 | vv = u.data{j};
|
---|
| 74 | aa = a(:,n+1:n+size(vv,1));
|
---|
| 75 | d = a(:,n+1:n+size(vv,1))*vv;
|
---|
| 76 | [dd,jj] = max(+d,[],2);
|
---|
| 77 | dd = zeros(size(dtot));
|
---|
| 78 | dd([1:m]'+(jj-1)*m) = alf(j);
|
---|
| 79 | dtot = dtot + dd;
|
---|
| 80 | n = n+size(vv,1);
|
---|
| 81 | end
|
---|
| 82 | elseif isstacked(u) % stacked combiner
|
---|
| 83 | for j=1:length(u.data) % weight them
|
---|
| 84 | vv = u.data{j};
|
---|
| 85 | d = a*vv;
|
---|
| 86 | [dd,jj] = max(+d,[],2);
|
---|
| 87 | dd = zeros(size(dtot));
|
---|
| 88 | dd([1:m]'+(jj-1)*m) = alf(j);
|
---|
| 89 | dtot = dtot + dd;
|
---|
| 90 | end
|
---|
| 91 | else
|
---|
| 92 | error('Classifier combination expected')
|
---|
| 93 | end
|
---|
| 94 | w = setdat(d,dtot);
|
---|
| 95 | end
|
---|
| 96 |
|
---|
| 97 | else % store classifier from given weights
|
---|
| 98 |
|
---|
| 99 | ismapping(a);
|
---|
| 100 | u = v; % the weights
|
---|
| 101 | v = a; % the combined classifier
|
---|
| 102 | n = length(v.data);
|
---|
| 103 | if length(u) ~= n
|
---|
| 104 | error('Wrong number of weights given')
|
---|
| 105 | end
|
---|
| 106 | [k,c] = getsize(v.data{1});
|
---|
| 107 | w = mapping(mfilename,'trained',{v,u},getlabels(v{1}),k,c);
|
---|
| 108 | w = setname(w,'Weighted Voting');
|
---|
| 109 | end
|
---|
| 110 |
|
---|
| 111 |
|
---|
| 112 | |
---|