%SETBATCH Enables/diasables execution in batch mode of mappings
%
% W = SETBATCH(W,FLAG,BATCHSIZE,OBJSIZE)
% W = W*SETBATCH([],FLAG,BATCHSIZE,OBJSIZE)
%
% INPUT
% W Mapping
% FLAG Switches batch processing on (TRUE, default) or off (FALSE)
% BATCHSIZE Number of objects per batch (default 1000)
% OBJSIZE Number of objects above which batch processing is enabled
% (default BATCHSIZE)
%
% OUTPUT
% W Mapping
%
%DESCRIPTION
%Mappings might be processed in batch mode in case of large datasets or
%datafiles. This may solve problems in computing A*W (A is a datase/file)
%by executing the result in batches if intermediate memory demand is too
%high. In case the final dataset is too large this will not be of use as it
%should fit in memory anyway.
%
%For some mappings batch processing is not possible as the processing of
%objects is not fully independent. For that reason it is skipped during
%training of mappings and classifiers.
%
%SEE ALSO (<a href="http://37steps.com/prtools">PRTools Guide</a>)
%MAPPINGS, GETBATCH
% Copyright: R.P.W. Duin, r.p.w.duin@37steps.com
function w = setbatch(varargin)
global DEFAULTBATCHSIZE
if isempty(DEFAULTBATCHSIZE)
DEFAULTBATCHSIZE = 1000;
end
argin = setdefaults(varargin,[],true,DEFAULTBATCHSIZE,[]);
[w,n,batchsize,objsize] = deal(argin{:});
if isempty(objsize), objsize = batchsize; end
if mapping_task(argin,'definition')
w = define_mapping(argin,'combiner');
w = setuser(w,{n,batchsize,objsize},'batch');
else
if isempty(objsize), objsize = batchsize; end
if all([0,1] ~= n)
error('Flag should be 0 or 1')
end
ismapping(w);
w = setuser(w,{n,batchsize,objsize},'batch');
end
return