Mercurial > hg > camir-aes2014

annotate toolboxes/MIRtoolbox1.3.2/somtoolbox/som_drmake.m @ 0:e9a9cd732c1e tip

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
first hg version after svn
author wolffd
date Tue, 10 Feb 2015 15:05:51 +0000
parents
children
rev   line source
wolffd@0 1 function [sR,best,sig,Cm] = som_drmake(D,inds1,inds2,sigmea,nanis)
wolffd@0 2
wolffd@0 3 % SOM_DRMAKE Make descriptive rules for given group within the given data.
wolffd@0 4 %
wolffd@0 5 % sR = som_drmake(D,[inds1],[inds2],[sigmea],[nanis])
wolffd@0 6 %
wolffd@0 7 % D (struct) map or data struct
wolffd@0 8 % (matrix) the data, of size [dlen x dim]
wolffd@0 9 % [inds1] (vector) indeces belonging to the group
wolffd@0 10 % (the whole data set by default)
wolffd@0 11 % [inds2] (vector) indeces belonging to the contrast group
wolffd@0 12 % (the rest of the data set by default)
wolffd@0 13 % [sigmea] (string) significance measure: 'accuracy',
wolffd@0 14 % 'mutuconf' (default), or 'accuracyI'.
wolffd@0 15 % (See definitions below).
wolffd@0 16 % [nanis] (scalar) value given for NaNs: 0 (=FALSE, default),
wolffd@0 17 % 1 (=TRUE) or NaN (=ignored)
wolffd@0 18 %
wolffd@0 19 % sR (struct array) best rule for each component. Each
wolffd@0 20 % struct has the following fields:
wolffd@0 21 % .type (string) 'som_rule'
wolffd@0 22 % .name (string) name of the component
wolffd@0 23 % .low (scalar) the low end of the rule range
wolffd@0 24 % .high (scalar) the high end of the rule range
wolffd@0 25 % .nanis (scalar) how NaNs are handled: NaN, 0 or 1
wolffd@0 26 %
wolffd@0 27 % best (vector) indeces of rules which make the best combined rule
wolffd@0 28 % sig (vector) significance measure values for each rule, and for the combined rule
wolffd@0 29 % Cm (matrix) A matrix of vectorized confusion matrices for each rule,
wolffd@0 30 % and for the combined rule: [a, c, b, d] (see below).
wolffd@0 31 %
wolffd@0 32 % For each rule, such rules sR.low <= x < sR.high are found
wolffd@0 33 % which optimize the given significance measure. The confusion
wolffd@0 34 % matrix below between the given grouping (G: group - not G: contrast group)
wolffd@0 35 % and rule (R: true or false) is used to determine the significance values:
wolffd@0 36 %
wolffd@0 37 % G not G
wolffd@0 38 % --------------- accuracy = (a+d) / (a+b+c+d)
wolffd@0 39 % true | a | b |
wolffd@0 40 % |-------------- mutuconf = a*a / ((a+b)(a+c))
wolffd@0 41 % false | c | d |
wolffd@0 42 % --------------- accuracyI = a / (a+b+c)
wolffd@0 43 %
wolffd@0 44 % See also SOM_DREVAL, SOM_DRTABLE.
wolffd@0 45
wolffd@0 46 % Contributed to SOM Toolbox 2.0, January 7th, 2002 by Juha Vesanto
wolffd@0 47 % Copyright (c) by Juha Vesanto
wolffd@0 48 % http://www.cis.hut.fi/projects/somtoolbox/
wolffd@0 49
wolffd@0 50 % Version 2.0beta juuso 070102
wolffd@0 51
wolffd@0 52 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
wolffd@0 53 %% input arguments
wolffd@0 54
wolffd@0 55 if isstruct(D),
wolffd@0 56 switch D.type,
wolffd@0 57 case 'som_data', cn = D.comp_names; D = D.data;
wolffd@0 58 case 'som_map', cn = D.comp_names; D = D.codebook;
wolffd@0 59 end
wolffd@0 60 else
wolffd@0 61 cn = cell(size(D,2),1);
wolffd@0 62 for i=1:size(D,2), cn{i} = sprintf('Variable%d',i); end
wolffd@0 63 end
wolffd@0 64
wolffd@0 65 [dlen,dim] = size(D);
wolffd@0 66 if nargin<2 | isempty(inds1), inds1 = 1:dlen; end
wolffd@0 67 if nargin<3 | isempty(inds2), i = ones(dlen,1); i(inds1) = 0; inds2 = find(i); end
wolffd@0 68 if nargin<4, sigmea = 'mutuconf'; end
wolffd@0 69 if nargin<5, nanis = 0; end
wolffd@0 70
wolffd@0 71 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
wolffd@0 72 %% input arguments
wolffd@0 73
wolffd@0 74 sig = zeros(dim+1,1);
wolffd@0 75 Cm = zeros(dim+1,4);
wolffd@0 76
wolffd@0 77 sR1tmp = struct('type','som_rule','name','','low',-Inf,'high',Inf,'nanis',nanis,'lowstr','','highstr','');
wolffd@0 78 sR = sR1tmp;
wolffd@0 79
wolffd@0 80 % single variable rules
wolffd@0 81 for i=1:dim,
wolffd@0 82
wolffd@0 83 % bin edges
wolffd@0 84 mi = min(D(:,i));
wolffd@0 85 ma = max(D(:,i));
wolffd@0 86 [histcount,bins] = hist([mi,ma],10);
wolffd@0 87 if size(bins,1)>1, bins = bins'; end
wolffd@0 88 edges = [-Inf, (bins(1:end-1)+bins(2:end))/2, Inf];
wolffd@0 89
wolffd@0 90 % find the rule for this variable
wolffd@0 91 [low,high,s,cm] = onevar_descrule(D(inds1,i),D(inds2,i),sigmea,nanis,edges);
wolffd@0 92 sR1 = sR1tmp;
wolffd@0 93 sR1.name = cn{i};
wolffd@0 94 sR1.low = low;
wolffd@0 95 sR1.high = high;
wolffd@0 96 sR(i) = sR1;
wolffd@0 97 sig(i) = s;
wolffd@0 98 Cm(i,:) = cm;
wolffd@0 99
wolffd@0 100 end
wolffd@0 101
wolffd@0 102 % find combined rule
wolffd@0 103 [dummy,order] = sort(-sig);
wolffd@0 104 maxsig = sig(order(1)); bestcm = Cm(order(1),:);
wolffd@0 105 best = order(1);
wolffd@0 106 for i=2:dim,
wolffd@0 107 com = [best, order(i)];
wolffd@0 108 [s,cm,truex,truey] = som_dreval(sR(com),D(:,com),sigmea,inds1,inds2,'and');
wolffd@0 109 if s>maxsig, best = com; maxsig = s; bestcm = cm; end
wolffd@0 110 end
wolffd@0 111 sig(end) = maxsig;
wolffd@0 112 Cm(end,:) = cm;
wolffd@0 113
wolffd@0 114 return;
wolffd@0 115
wolffd@0 116 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%55
wolffd@0 117 %% descriptive rules
wolffd@0 118
wolffd@0 119 function [low,high,sig,cm] = onevar_descrule(x,y,sigmea,nanis,edges)
wolffd@0 120
wolffd@0 121 % Given a set of bin edges, find the range of bins with best significance.
wolffd@0 122 %
wolffd@0 123 % x data values in cluster
wolffd@0 124 % y data values not in cluster
wolffd@0 125 % sigmea significance measure
wolffd@0 126 % bins bin centers
wolffd@0 127 % nanis how to handle NaNs
wolffd@0 128
wolffd@0 129 % histogram counts
wolffd@0 130 if isnan(nanis), x = x(~isnan(x)); y = y(~isnan(y)); end
wolffd@0 131 [xcount,xbin] = histc(x,edges);
wolffd@0 132 [ycount,ybin] = histc(y,edges);
wolffd@0 133 xcount = xcount(1:end-1);
wolffd@0 134 ycount = ycount(1:end-1);
wolffd@0 135 xnan=sum(isnan(x));
wolffd@0 136 ynan=sum(isnan(y));
wolffd@0 137
wolffd@0 138 % find number of true items in both groups in all possible ranges
wolffd@0 139 n = length(xcount);
wolffd@0 140 V = zeros(n*(n+1)/2,4);
wolffd@0 141 s1 = cumsum(xcount);
wolffd@0 142 s2 = cumsum(xcount(end:-1:1)); s2 = s2(end:-1:1);
wolffd@0 143 m = s1(end);
wolffd@0 144 Tx = triu(s1(end)-m*log(exp(s1/m)*exp(s2/m)')+repmat(xcount',[n 1])+repmat(xcount,[1 n]),0);
wolffd@0 145 s1 = cumsum(ycount);
wolffd@0 146 s2 = cumsum(ycount(end:-1:1)); s2 = s2(end:-1:1);
wolffd@0 147 Ty = triu(s1(end)-m*log(exp(s1/m)*exp(s2/m)')+repmat(ycount',[n 1])+repmat(ycount,[1 n]),0);
wolffd@0 148 [i,j] = find(Tx+Ty);
wolffd@0 149 k = sub2ind(size(Tx),i,j);
wolffd@0 150 V = [i, j, Tx(k), Ty(k)];
wolffd@0 151 tix = V(:,3) + nanis*xnan;
wolffd@0 152 tiy = V(:,4) + nanis*ynan;
wolffd@0 153
wolffd@0 154 % select the best range
wolffd@0 155 nix = length(x);
wolffd@0 156 niy = length(y);
wolffd@0 157 Cm = [tix,nix-tix,tiy,niy-tiy];
wolffd@0 158 [s,k] = max(som_drsignif(sigmea,Cm));
wolffd@0 159
wolffd@0 160 % output
wolffd@0 161 low = edges(V(k,1));
wolffd@0 162 high = edges(V(k,2)+1);
wolffd@0 163 sig = s;
wolffd@0 164 cm = Cm(k,:);
wolffd@0 165
wolffd@0 166 return;
wolffd@0 167