fitnonlinearmodel_consolidate

 function fitnonlinearmodel_consolidate(dir0)

 <dir0> is a directory containing results from fitnonlinearmodel.m

 load in all of the *.mat files in <dir0> and write 
 the consolidated results to <dir0>.mat.

 we assume that the full set of voxels have been analyzed
 (in one or more chunks).  only the primary outputs of
 fitnonlinearmodel.m are saved to the new file; the auxiliary
 outputs (that exist in individual .mat files) are not.

 note that we check to make sure that the total number of
 voxels that are found in the .mat files is equal to the total
 number of voxels specified in the original call to fitnonlinearmodel.m.
 this ensures that all voxels have been analyzed!

 example:

 % first, set up the problem
 x = randn(100,1);
 y = bsxfun(@plus,2*x + 3,randn(100,20));
 opt = struct( ...
   'outputdir','test', ...
   'stimulus',[x ones(100,1)], ...
   'data',y, ...
   'model',{{[1 1] [-Inf -Inf; Inf Inf] @(pp,dd) dd*pp'}});

 % next, do the fitting in chunks of 2
 for p=1:10
   fitnonlinearmodel(opt,2,p);
 end

 % then, consolidate the results
 fitnonlinearmodel_consolidate('test');

 % check the output
 a = load('test.mat');
 a

0001 function fitnonlinearmodel_consolidate(dir0)
0002 
0003 % function fitnonlinearmodel_consolidate(dir0)
0004 %
0005 % <dir0> is a directory containing results from fitnonlinearmodel.m
0006 %
0007 % load in all of the *.mat files in <dir0> and write
0008 % the consolidated results to <dir0>.mat.
0009 %
0010 % we assume that the full set of voxels have been analyzed
0011 % (in one or more chunks).  only the primary outputs of
0012 % fitnonlinearmodel.m are saved to the new file; the auxiliary
0013 % outputs (that exist in individual .mat files) are not.
0014 %
0015 % note that we check to make sure that the total number of
0016 % voxels that are found in the .mat files is equal to the total
0017 % number of voxels specified in the original call to fitnonlinearmodel.m.
0018 % this ensures that all voxels have been analyzed!
0019 %
0020 % example:
0021 %
0022 % % first, set up the problem
0023 % x = randn(100,1);
0024 % y = bsxfun(@plus,2*x + 3,randn(100,20));
0025 % opt = struct( ...
0026 %   'outputdir','test', ...
0027 %   'stimulus',[x ones(100,1)], ...
0028 %   'data',y, ...
0029 %   'model',{{[1 1] [-Inf -Inf; Inf Inf] @(pp,dd) dd*pp'}});
0030 %
0031 % % next, do the fitting in chunks of 2
0032 % for p=1:10
0033 %   fitnonlinearmodel(opt,2,p);
0034 % end
0035 %
0036 % % then, consolidate the results
0037 % fitnonlinearmodel_consolidate('test');
0038 %
0039 % % check the output
0040 % a = load('test.mat');
0041 % a
0042 
0043 % consolidate
0044 file0 = consolidatematdir(dir0,{'opt'});  % 'opt' may be big. let's specifically exclude it.
0045 
0046 % load
0047 a = load(file0);
0048 
0049 % what is the total number of voxels (so we can check)
0050 totalnumvxs = a.results(1).totalnumvxs;
0051 
0052 % assign to b, consolidating as we go
0053 clear b;
0054 varlist = {'params' 'trainperformance' 'testperformance' 'aggregatedtestperformance' 'testdata' 'modelpred' 'modelfit' 'numiters' 'resnorms'};
0055 dimlist = [3 2 2 2 2 2 3 2 2];
0056 for zz=1:length(varlist)
0057   if isfield(a.results(1),varlist{zz});
0058      
0059     % cat the results
0060     temp = cat(dimlist(zz),a.results.(varlist{zz}));
0061 
0062     % check if we have the full set of results.  note that sometimes outputs can be empty.
0063     assert(isempty(temp) || size(temp,dimlist(zz)) == totalnumvxs, ...
0064            'we did not find the full set of results!');
0065     
0066     % record
0067     b.(varlist{zz}) = temp;
0068 
0069   end
0070 end
0071 
0072 % save
0073 save(file0,'-struct','b');