chunkfun

 function f = chunkfun(v,nums,fun)

 <v> is a row or column vector
 <nums> is a vector of counts such that sum(<nums>) is equal to length(<v>)
 <fun> (optional) is a function that takes A x B to 1 x B.  each column consists
   of a mix of valid elements (comprising a group in <v>) and NaNs.  <fun>
   should compute something based on the valid elements in each column.
   default: @(y)nansum(y,1).

 return a row vector by applying <fun> to <v> reshaped into groups.
 we emphasize fast execution!

 example:
 isequal(chunkfun([1 2 3 5 6],[3 2]),[6 11])

0001 function f = chunkfun(v,nums,fun)
0002 
0003 % function f = chunkfun(v,nums,fun)
0004 %
0005 % <v> is a row or column vector
0006 % <nums> is a vector of counts such that sum(<nums>) is equal to length(<v>)
0007 % <fun> (optional) is a function that takes A x B to 1 x B.  each column consists
0008 %   of a mix of valid elements (comprising a group in <v>) and NaNs.  <fun>
0009 %   should compute something based on the valid elements in each column.
0010 %   default: @(y)nansum(y,1).
0011 %
0012 % return a row vector by applying <fun> to <v> reshaped into groups.
0013 % we emphasize fast execution!
0014 %
0015 % example:
0016 % isequal(chunkfun([1 2 3 5 6],[3 2]),[6 11])
0017 
0018 % input
0019 if ~exist('fun','var') || isempty(fun)
0020   fun = @(y)nansum(y,1);
0021 end
0022 
0023 % do it
0024 f = NaN(max(nums),length(nums));
0025 f(bsxfun(@le,repmat((1:max(nums))',[1 length(nums)]),nums)) = v;
0026 f = feval(fun,f);