annotate solvers/SMALL_chol.m @ 104:e2ce05e21a55

Merge
author vemiya <valentin.emiya@inria.fr>
date Tue, 12 Apr 2011 16:11:41 +0200
parents f6cedfec9ffb
children 8e660fd14774
rev   line source
idamnjanovic@1 1 function [A]=SMALL_chol(Dict,X, m, maxNumCoef, errorGoal, varargin)
idamnjanovic@1 2 %%
idamnjanovic@22 3 %
idamnjanovic@22 4 % Centre for Digital Music, Queen Mary, University of London.
idamnjanovic@22 5 % This file copyright 2009 Ivan Damnjanovic.
idamnjanovic@22 6 %
idamnjanovic@22 7 % This program is free software; you can redistribute it and/or
idamnjanovic@22 8 % modify it under the terms of the GNU General Public License as
idamnjanovic@22 9 % published by the Free Software Foundation; either version 2 of the
idamnjanovic@22 10 % License, or (at your option) any later version. See the file
idamnjanovic@22 11 % COPYING included with this distribution for more information.
idamnjanovic@22 12 %
idamnjanovic@1 13 % Sparse coding of a group of signals based on a given
idamnjanovic@1 14 % dictionary and specified number of atoms to use.
idamnjanovic@1 15 % input arguments: Dict - the dictionary
idamnjanovic@1 16 % X - the signals to represent
idamnjanovic@1 17 % m - number of atoms in Dictionary
idamnjanovic@1 18 % errorGoal - the maximal allowed representation error for
idamnjanovic@1 19 % each signal.
idamnjanovic@1 20 %
idamnjanovic@1 21 % optional: if Dict is function handle then Transpose Dictionary
idamnjanovic@1 22 % handle needs to be specified.
idamnjanovic@1 23 %
idamnjanovic@1 24 % output arguments: A - sparse coefficient matrix.
idamnjanovic@1 25 %
idamnjanovic@1 26 %%
idamnjanovic@1 27 % This Dictionary check is based on Thomas Blumensath work in sparsify 0_4 greedy solvers
idamnjanovic@1 28 explicitD=0;
idamnjanovic@1 29 if isa(Dict,'float')
idamnjanovic@1 30 D =@(z) Dict*z;
idamnjanovic@1 31 Dt =@(z) Dict'*z;
idamnjanovic@63 32 explicitD=0;
idamnjanovic@1 33 elseif isobject(Dict)
idamnjanovic@1 34 D =@(z) Dict*z;
idamnjanovic@1 35 Dt =@(z) Dict'*z;
idamnjanovic@1 36 elseif isa(Dict,'function_handle')
idamnjanovic@1 37 try
idamnjanovic@1 38 DictT=varargin{1};
idamnjanovic@1 39 if isa(DictT,'function_handle');
idamnjanovic@74 40 D=Dict;
idamnjanovic@74 41 Dt=DictT;
idamnjanovic@1 42 else
idamnjanovic@1 43 error('If Dictionary is a function handle,Transpose Dictionary also needs to be a function handle. ');
idamnjanovic@1 44 end
idamnjanovic@1 45 catch
idamnjanovic@1 46 error('If Dictionary is a function handle, Transpose Dictionary needs to be specified. Exiting.');
idamnjanovic@1 47 end
idamnjanovic@1 48 else
idamnjanovic@1 49 error('Dictionary is of unsupported type. Use explicit matrix, function_handle or object. Exiting.');
idamnjanovic@1 50 end
idamnjanovic@1 51 %%
idamnjanovic@1 52 [n,P]=size(X);
idamnjanovic@1 53
idamnjanovic@1 54
idamnjanovic@1 55
idamnjanovic@1 56 global opts opts_tr machPrec
idamnjanovic@1 57 opts.UT = true;
idamnjanovic@1 58 opts_tr.UT = true; opts_tr.TRANSA = true;
idamnjanovic@1 59 machPrec = 1e-5;
idamnjanovic@1 60
idamnjanovic@1 61 A = sparse(m,size(X,2));
idamnjanovic@1 62 for k=1:1:P,
idamnjanovic@1 63
idamnjanovic@1 64 R_I = [];
idamnjanovic@1 65 x=X(:,k);
idamnjanovic@1 66 residual=x;
idamnjanovic@1 67 indx = [];
idamnjanovic@1 68 a = zeros(m,1);
idamnjanovic@1 69 currResNorm = norm(residual);
idamnjanovic@63 70 errorGoal1=errorGoal*currResNorm;
idamnjanovic@1 71 j = 0;
idamnjanovic@1 72 while currResNorm>errorGoal & j < maxNumCoef,
idamnjanovic@1 73 j = j+1;
idamnjanovic@1 74 dir=Dt(residual);
idamnjanovic@1 75
idamnjanovic@1 76 [tmp__, pos]=max(abs(dir));
idamnjanovic@1 77
idamnjanovic@1 78 [R_I, flag] = updateChol(R_I, n, m, D, explicitD, indx, pos, Dt);
idamnjanovic@1 79
idamnjanovic@1 80
idamnjanovic@1 81 indx(j)=pos;
idamnjanovic@1 82 dx=zeros(m,1);
idamnjanovic@1 83
idamnjanovic@1 84 z = linsolve(R_I,dir(indx),opts_tr);
idamnjanovic@1 85
idamnjanovic@1 86 dx(indx) = linsolve(R_I,z,opts);
idamnjanovic@1 87 a(indx) = a(indx) + dx(indx);
idamnjanovic@1 88
idamnjanovic@1 89 residual=x-D(a);
idamnjanovic@1 90 currResNorm = norm(residual);
idamnjanovic@1 91
idamnjanovic@1 92
idamnjanovic@1 93 end;
idamnjanovic@1 94 if (~isempty(indx))
idamnjanovic@1 95 A(indx,k)=a(indx);
idamnjanovic@1 96 end
idamnjanovic@1 97 end;
idamnjanovic@1 98 return;
idamnjanovic@1 99
idamnjanovic@1 100
idamnjanovic@1 101 function [R, flag] = updateChol(R, n, N, A, explicitA, activeSet, newIndex, varargin)
idamnjanovic@1 102
idamnjanovic@1 103 % updateChol: Updates the Cholesky factor R of the matrix
idamnjanovic@1 104 % A(:,activeSet)'*A(:,activeSet) by adding A(:,newIndex)
idamnjanovic@1 105 % If the candidate column is in the span of the existing
idamnjanovic@1 106 % active set, R is not updated, and flag is set to 1.
idamnjanovic@1 107
idamnjanovic@1 108 global opts_tr machPrec
idamnjanovic@1 109 flag = 0;
idamnjanovic@1 110
idamnjanovic@1 111 if (explicitA)
idamnjanovic@1 112 newVec = A(:,newIndex);
idamnjanovic@1 113 else
idamnjanovic@1 114 At=varargin{1};
idamnjanovic@1 115 e = zeros(N,1);
idamnjanovic@1 116 e(newIndex) = 1;
idamnjanovic@1 117 newVec = A(e);%feval(A,1,n,N,e,1:N,N);
idamnjanovic@1 118 end
idamnjanovic@1 119
idamnjanovic@1 120 if isempty(activeSet),
idamnjanovic@1 121 R = sqrt(sum(newVec.^2));
idamnjanovic@1 122 else
idamnjanovic@1 123 if (explicitA)
idamnjanovic@1 124 p = linsolve(R,A(:,activeSet)'*A(:,newIndex),opts_tr);
idamnjanovic@1 125 else
idamnjanovic@1 126 AnewVec = At(newVec);%feval(A,2,n,length(activeSet),newVec,activeSet,N);
idamnjanovic@1 127 p = linsolve(R,AnewVec(activeSet),opts_tr);
idamnjanovic@1 128 end
idamnjanovic@1 129 q = sum(newVec.^2) - sum(p.^2);
idamnjanovic@1 130 if (q <= machPrec) % Collinear vector
idamnjanovic@1 131 flag = 1;
idamnjanovic@1 132 else
idamnjanovic@1 133 R = [R p; zeros(1, size(R,2)) sqrt(q)];
idamnjanovic@1 134 end
idamnjanovic@1 135 end