why arrayfun does NOT improve my struct array operation performance

here is the input data:

% @param Landmarks:
%           Landmarks should be 1*m struct. 
%           m is the number of training set.
%           Landmark(i).data is a n*2 matrix

old function:

function Landmarks=CenterOfGravity(Landmarks)
% align center of gravity

for i=1 : length(Landmarks)
    Landmarks(i).data=Landmarks(i).data - ones(size(Landmarks(i).data,1),1)...
        *mean(Landmarks(i).data);
end
end

new function which use arrayfun:

function [Landmarks] = center_to_gravity(Landmarks)
Landmarks = arrayfun(@(struct_data)...
                        struct('data', struct_data.data - repmat(mean(struct_data.data), [size(struct_data.data, 1), 1]))...
                                            ,Landmarks);
end %function center_to_gravity

when using profiler, I find the usage of time is NOT what I expected:

Function          Total Time    Self Time*
CenterOfGravity     0.011s  0.004 s
center_to_gravity   0.029s  0.001 s

Can someone tell me why?

Answer 1

[Jan]

ARRAYFUN is not more efficient than a FOR loop, because it has a FOR loop internally.

Another idea:

for i = 1 : length(Landmarks)
    data = Landmarks(i).data;
    Landmarks(i).data= bsxfun(@minus, data, sum(data, 1) / size(data, 1));
end

Reduce the repeated access to a field, SUM/LENGTH is faster than MEAN, BSXFUN avoid the creation of a temporary array.

Or:

for i = 1 : length(Landmarks)
    data = Landmarks(i).data;
    m = sum(data, 1) / size(data, 1);
    data(:, 1) = data(:, 1) - m(1);
    data(:, 2) = data(:, 2) - m(2);
    Landmarks(i).data = data;
end

[HaveF]

Surprisingly, the second approach is fast then then first(test with timeit).

[HaveF]

In terms of why, Pursuit's explanation seems reasonable.

[HaveF]

http://stackoverflow.com/questions/11170089/why-arrayfun-does-not-improve-my-struct-array-operation-performance

Answer 2

[Walter Roberson]

Also, profile disables a number of optimizations, so you cannot use profiler to determine full execution rate. Try timeit

[HaveF]

Thanks for this great tip! I decide to use timeit when compare two methods, use profiler to determine the bottleneck of my program for its relative inaccuracy time.

[HaveF]

BTW, when I use profiler, I have to set the number of active CPUs to one before my start profiling, should I do this before I use timeit?