svmlwrite.m

function svmlwrite(fname, X, Y, floatformat)
% SVMLWRITE - Write matrix into data file for SVM light
% 
%   SVMLWRITE(FNAME, X) writes out matrix X into file FNAME, in the format
%   needed for SVM light. X is a matrix of data points, with one point
%   per row.
%   SVMLWRITE(FNAME, X, Y) writes out data points X with target values
%   given in the column vector Y. In the case of classification, Y(i) may
%   either be +1, -1 or 0 (indicating unknown class label, for
%   transductive SVM). In the case of regression data, Y(i) is
%   real-valued.
%   SVMLWRITE(FNAME, X, Y, FLOATFORMAT) uses format string FLOATFORMAT to
%   write out the features in X (and in case of regression, also the
%   target values Y(i)). Default: '%.16g'
%
%   See also
%   SVM_LEARN, SVMLOPT, SVM_CLASSIFY, SVMLREAD
%

% 
% Copyright (c) by Anton Schwaighofer (2001)
% $Revision: 1.8 $ $Date: 2002/02/19 12:28:03 $
% mailto:anton.schwaighofer@gmx.net
% 
% This program is released unter the GNU General Public License.
% 

error(nargchk(2, 4, nargin));
if nargin<4,
  floatformat = '%.16g';
end
[N, d] = size(X);
if nargin<3,
  Y = zeros(N, 1);
end
if isempty(Y),
  Y = zeros(N, 1);
end
if ~all(size(Y)==[N 1]),
  error('Input parameter Y must be a column vector with length SIZE(X,1)');
end

f = fopen(fname, 'wt');
if (f<0),
  error(sprintf('Unable to open file %s', fname));
end

% Check whether this is a regression or a classification problem
uY = unique(Y);
if isempty(setdiff(uY, [-1 0 +1])),
  % Classification:
  labelformat = '%i ';
else
  % Regression: 
  labelformat = [floatformat ' '];
end
% transpose for increased efficiency when working with sparse matrices
X = X';
fprintf('Writing ');
for i = 1:N,
  Xi = X(:,i);
  % Write label as the first entry
  s = sprintf(labelformat, Y(i));
  % Then follow 'feature:value' pairs
  ind = find(Xi);
  s = [s sprintf(['%i:' floatformat ' '], [ind'; full(Xi(ind))'])];
  fprintf(f, '%s\n', s);
  if (rem(i,100)==0),
    fprintf(' %i', i);
  end
end
fprintf(' done.\n');

fclose(f);