Mercurial > hg > camir-aes2014

annotate toolboxes/MIRtoolbox1.3.2/somtoolbox/som_stats.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 csS = som_stats(D,varargin)
wolffd@0 2
wolffd@0 3 %SOM_STATS Calculate descriptive statistics for the data.
wolffd@0 4 %
wolffd@0 5 % csS = som_stats(D,[sort]);
wolffd@0 6 %
wolffd@0 7 % csS = som_stats(D);
wolffd@0 8 % csS = som_stats(D,'nosort');
wolffd@0 9 % som_table_print(som_stats_table(csS))
wolffd@0 10 %
wolffd@0 11 % Input and output arguments ([]'s are optional):
wolffd@0 12 % D (matrix) a matrix, size dlen x dim
wolffd@0 13 % (struct) data or map struct
wolffd@0 14 % [sort] (string) 'sort' (default) or 'nosort'
wolffd@0 15 % If 'nosort' is specified, the data is not
wolffd@0 16 % sorted, and therefore the values of
wolffd@0 17 % nunique, uvalues, ucount, fvalues, fcount, and tiles fields
wolffd@0 18 % are not calculated. This may be useful if
wolffd@0 19 % there is a very large amount of data, and
wolffd@0 20 % one wants to reduce calculation time.
wolffd@0 21 %
wolffd@0 22 % csS (cell array) size dim x 1, of statistics structs with
wolffd@0 23 % the following fields
wolffd@0 24 % .type (string) 'som_stat'
wolffd@0 25 % .name (string) name of the variable
wolffd@0 26 % .normalization (struct array) variable normalization (see SOM_NORMALIZE)
wolffd@0 27 % .ntotal (scalar) total number of values
wolffd@0 28 % .nvalid (scalar) number of valid values (not Inf or NaN)
wolffd@0 29 % .min (scalar) minimum value
wolffd@0 30 % .max (scalar) maximum value
wolffd@0 31 % .mean (scalar) mean value (not Inf or NaN)
wolffd@0 32 % .std (scalar) standard deviation (not Inf or NaN)
wolffd@0 33 % .nunique (scalar) number of unique values
wolffd@0 34 % .mfvalue (vector) most frequent value
wolffd@0 35 % .mfcount (vector) number of occurances of most frequent value
wolffd@0 36 % .values (vector) at most MAXDISCRETE (see below) sample values
wolffd@0 37 % .counts (vector) number of occurances for each sampled value
wolffd@0 38 % .tiles (vector) NT-tile values, for example
wolffd@0 39 % NT=4 for quartiles: 25%, 50% and 75%
wolffd@0 40 % NT=100 for percentiles: 1%, 2%, ... and 99%
wolffd@0 41 % .hist (struct) histogram struct with the following fields
wolffd@0 42 % .type (string) 'som_hist'
wolffd@0 43 % .bins (vector) histogram bin centers
wolffd@0 44 % .counts (vector) count of values in each bin
wolffd@0 45 % .binlabels (cellstr) labels for the bins (denormalized bin
wolffd@0 46 % center values)
wolffd@0 47 % .binlabels2 (cellstr) labels for the bins (denormalized bin
wolffd@0 48 % edge values, e.g. '[1.4,2.5['
wolffd@0 49 %
wolffd@0 50 % Constants:
wolffd@0 51 % MAXDISCRETE = 10
wolffd@0 52 % NT = 10
wolffd@0 53 %
wolffd@0 54 % See also SOM_STATS_PLOT, SOM_STATS_TABLE, SOM_TABLE_PRINT, SOM_STATS_REPORT.
wolffd@0 55
wolffd@0 56 % Contributed to SOM Toolbox 2.0, December 31st, 2001 by Juha Vesanto
wolffd@0 57 % Copyright (c) by Juha Vesanto
wolffd@0 58 % http://www.cis.hut.fi/projects/somtoolbox/
wolffd@0 59
wolffd@0 60 % Version 2.0beta juuso 311201
wolffd@0 61
wolffd@0 62 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%5
wolffd@0 63 %% arguments
wolffd@0 64
wolffd@0 65 % default values
wolffd@0 66 nosort = 0;
wolffd@0 67 nbins = 10;
wolffd@0 68 maxdiscrete = 20;
wolffd@0 69 ntiles = 10;
wolffd@0 70
wolffd@0 71 % first argument
wolffd@0 72 if isstruct(D),
wolffd@0 73 switch D.type,
wolffd@0 74 case 'som_map', cn = D.comp_names; sN = D.comp_norm; D = D.codebook;
wolffd@0 75 case 'som_data', cn = D.comp_names; sN = D.comp_norm; D = D.data;
wolffd@0 76 otherwise, error('Invalid first argument')
wolffd@0 77 end
wolffd@0 78 else
wolffd@0 79 cn = cell(size(D,2),1);
wolffd@0 80 cn(:) = {'Variable'};
wolffd@0 81 for i=1:length(cn), cn{i} = sprintf('%s%d',cn{i},i); end
wolffd@0 82 sN = cell(size(D,2),1);
wolffd@0 83 end
wolffd@0 84 [dlen dim] = size(D);
wolffd@0 85
wolffd@0 86 % other arguments
wolffd@0 87
wolffd@0 88 if length(varargin)>0,
wolffd@0 89 if strcmp(varargin{1},'nosort'), nosort = 1; end
wolffd@0 90 end
wolffd@0 91
wolffd@0 92 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%5
wolffd@0 93 %% action
wolffd@0 94
wolffd@0 95 sStat = struct('type','som_stat','name','','normalization',[],...
wolffd@0 96 'min',NaN,'max',NaN,'mean',NaN,'std',NaN,...
wolffd@0 97 'nunique',NaN,'values',[],'counts',[],'mfvalue',NaN,'mfcount',NaN,'tiles',[],...
wolffd@0 98 'ntotal',dlen,'nvalid',NaN,'hist',[]);
wolffd@0 99 csS = cell(0);
wolffd@0 100
wolffd@0 101 for i=1:dim,
wolffd@0 102 sS = sStat;
wolffd@0 103 sS.name = cn{i};
wolffd@0 104 sS.normalization = sN{i};
wolffd@0 105 x = D(:,i);
wolffd@0 106 x(find(~isfinite(x))) = [];
wolffd@0 107 % basic descriptive statistics
wolffd@0 108 sS.nvalid = length(x);
wolffd@0 109 if length(x),
wolffd@0 110 sS.min = min(x);
wolffd@0 111 sS.max = max(x);
wolffd@0 112 sS.mean = mean(x);
wolffd@0 113 sS.std = std(x);
wolffd@0 114 bins = [];
wolffd@0 115 if ~nosort,
wolffd@0 116 xsorted = sort(x);
wolffd@0 117 % number of unique values
wolffd@0 118 repeated = (xsorted(1:end-1)==xsorted(2:end));
wolffd@0 119 j = [1; find(~repeated)+1];
wolffd@0 120 xunique = xsorted(j);
wolffd@0 121 sS.nunique = length(xunique);
wolffd@0 122 ucount = diff([j; length(xsorted)+1]);
wolffd@0 123 % most frequent value
wolffd@0 124 [fcount,j] = max(ucount);
wolffd@0 125 sS.mfvalue = xunique(j);
wolffd@0 126 sS.mfcount = fcount;
wolffd@0 127 % -tiles (k*100/ntiles % of values, k=1..)
wolffd@0 128 pickind = round(linspace(1,sS.nvalid,ntiles+1));
wolffd@0 129 pickind = pickind(2:end-1);
wolffd@0 130 sS.tiles = xsorted(pickind);
wolffd@0 131 if sS.nunique <= sS.nvalid/2,
wolffd@0 132 % unique values
wolffd@0 133 sS.values = xunique;
wolffd@0 134 sS.counts = ucount;
wolffd@0 135 bins = sS.values;
wolffd@0 136 else
wolffd@0 137 % just maxdiscrete values, evenly picked
wolffd@0 138 pickind = round(linspace(1,sS.nunique,maxdiscrete));
wolffd@0 139 sS.values = xunique(pickind);
wolffd@0 140 sS.counts = ucount(pickind);
wolffd@0 141
wolffd@0 142 %% OPTION 2: maxdiscrete most frequent values
wolffd@0 143 %[v,j] = sort(ucount);
wolffd@0 144 %pickind = j(1:maxdiscrete);
wolffd@0 145 %sS.values = xunique(pickind);
wolffd@0 146 %sS.counts = ucount(pickind);
wolffd@0 147
wolffd@0 148 % OPTION 3: representative values - calculated using k-means
wolffd@0 149 %[y,bm,qe] = kmeans(x,maxdiscrete);
wolffd@0 150 %sS.values = y;
wolffd@0 151 %sS.counts = full(sum(sparse(bm,1:length(bm),1,maxdiscrete,length(bm)),2));
wolffd@0 152 end
wolffd@0 153 end
wolffd@0 154 if isempty(bins),
wolffd@0 155 bins = linspace(sS.min,sS.max,nbins+1);
wolffd@0 156 bins = (bins(1:end-1)+bins(2:end))/2;
wolffd@0 157 end
wolffd@0 158 sS.hist = som_hist(x,bins,sS.normalization);
wolffd@0 159 else
wolffd@0 160 sS.hist = som_hist(x,0);
wolffd@0 161 end
wolffd@0 162 csS{end+1} = sS;
wolffd@0 163 end
wolffd@0 164
wolffd@0 165 return;
wolffd@0 166
wolffd@0 167 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%5
wolffd@0 168 %% subfunctions
wolffd@0 169
wolffd@0 170 function sH = som_hist(x,bins,sN)
wolffd@0 171
wolffd@0 172 binlabels = [];
wolffd@0 173 binlabels2 = [];
wolffd@0 174 if nargin<2 | isempty(bins) | isnan(bins),
wolffd@0 175 bins = linspace(min(x),max(x),10);
wolffd@0 176 end
wolffd@0 177 if isstruct(bins),
wolffd@0 178 bins = sH.bins;
wolffd@0 179 binlabels = sH.binlabels;
wolffd@0 180 binlabels2 = sH.binlabels2;
wolffd@0 181 end
wolffd@0 182 if nargin<3, sN = []; end
wolffd@0 183
wolffd@0 184 sH = struct('type','som_hist','bins',bins,'counts',[],...
wolffd@0 185 'binlabels',binlabels,'binlabels2',binlabels2);
wolffd@0 186
wolffd@0 187 if length(bins)==1,
wolffd@0 188 sH.counts = [length(x)];
wolffd@0 189 edges = bins;
wolffd@0 190 elseif length(x),
wolffd@0 191 edges = (bins(1:end-1)+bins(2:end))/2;
wolffd@0 192 counts = histc(x,[-Inf; edges(:); Inf]);
wolffd@0 193 sH.counts = counts(1:end-1);
wolffd@0 194 end
wolffd@0 195
wolffd@0 196 if isempty(sH.binlabels),
wolffd@0 197 b = som_denormalize(bins(:),sN);
wolffd@0 198 sH.binlabels = numtostring(b,4);
wolffd@0 199 end
wolffd@0 200
wolffd@0 201 if isempty(sH.binlabels2),
wolffd@0 202 if length(edges)==1,
wolffd@0 203 sH.binlabels2 = numtostring(som_denormalize(edges,sN),2);
wolffd@0 204 if length(bins)>1,
wolffd@0 205 sH.binlabels2 = sH.binlabels2([1 1]);
wolffd@0 206 sH.binlabels2{1} = [']' sH.binlabels2{1} '['];
wolffd@0 207 sH.binlabels2{2} = ['[' sH.binlabels2{2} '['];
wolffd@0 208 end
wolffd@0 209 else
wolffd@0 210 if size(edges,1)==1, edges = edges'; end
wolffd@0 211 bstr = numtostring(som_denormalize(edges,sN),4);
wolffd@0 212 sH.binlabels2 = bstr([1:end end]);
wolffd@0 213 sH.binlabels2{1} = [bstr{1} '['];
wolffd@0 214 for i=2:length(sH.binlabels2)-1,
wolffd@0 215 sH.binlabels2{i} = ['[' bstr{i-1} ',' bstr{i} '['];
wolffd@0 216 end
wolffd@0 217 sH.binlabels2{end} = ['[' bstr{end}];
wolffd@0 218 end
wolffd@0 219 end
wolffd@0 220
wolffd@0 221 if 0,
wolffd@0 222 if length(bins)==1, sH.binlabels2 = {'constant'};
wolffd@0 223 else
wolffd@0 224 ntiles = 10;
wolffd@0 225 plim = [1:ntiles-1] / ntiles;
wolffd@0 226 cp = cumsum(sH.counts)/sum(sH.counts);
wolffd@0 227 [dummy,i] = histc(cp,[-Inf plim Inf]);
wolffd@0 228 l2 = cell(length(bins),1);
wolffd@0 229 for j=1:length(bins), l2{j} = sprintf('Q%d',i(j)); end
wolffd@0 230 if i(1) > 1, l2{1} = ['...' l2{1}]; end
wolffd@0 231 k = 0;
wolffd@0 232 for j=2:length(bins),
wolffd@0 233 if i(j)==i(j-1),
wolffd@0 234 if k==0, l2{j-1} = [l2{j-1} '.1']; k = 1; end
wolffd@0 235 k = k + 1;
wolffd@0 236 l2{j} = [l2{j} '.' num2str(k)];
wolffd@0 237 else k = 0; end
wolffd@0 238 end
wolffd@0 239 if i(end) < ntiles, l2{end} = [l2{end} '...']; end
wolffd@0 240 sH.binlabels2 = l2;
wolffd@0 241 end
wolffd@0 242 end
wolffd@0 243
wolffd@0 244 return;
wolffd@0 245
wolffd@0 246 function vstr = numtostring(v,d)
wolffd@0 247
wolffd@0 248 r = max(v)-min(v);
wolffd@0 249 if r==0, r=1; end
wolffd@0 250 nearzero = (abs(v)/r < 10.^-d);
wolffd@0 251 i1 = find(v > 0 & nearzero);
wolffd@0 252 i2 = find(v < 0 & nearzero);
wolffd@0 253 vstr = strrep(cellstr(num2str(v,d)),' ','');
wolffd@0 254 vstr(i1) = {'0.0'};
wolffd@0 255 vstr(i2) = {'-0.0'};
wolffd@0 256 return;
wolffd@0 257