Preallocating for speed, what kind of variable is this?

I'm looking through some code and I came across this loop where Box should be preallocated for speed. Box, is the variable that I should preallocate outside the loop. But I don't know what kind of a variable it is. Can anyone help by looking at this?

for posi = 1:Length
  nArea=nBox(posi).Area;    % Area of number/letters
  nBox(posi).BoundingBox;         % Frame objects
  if nArea>50 && nArea<200 && ...
       nBox(posi).BoundingBox(4)>10 && nBox(posi).BoundingBox(3)>7 && ...
       nBox(posi).BoundingBox(3)<25
    k=k+1;
    Box(k,:)=round(nBox(posi).BoundingBox); 
  end
end

Answer 1

[dpb]

Box is just a 2D array of type whatever is nBox.BoundingBox (which would appear to be a floating point type given the round operation, else't it's a waste).

As for preallocating, the size isn't known a priori; only the number of rows that satisfy the condition will be allocated. Depending on the matnitude of the upper bound of the loop Length, the time savings may be trivial compared to the overhead of allocating a set, checking for overrunning and reallocating, etc., etc., etc., ...

You could, of course, allocate for the full Length rows and clear k+1:end after the loop finishes. The 2nd dimension is whatever is the size of the structure array.

It appears that the loop could be replaced if this is all there is by --

    idx= isbetween(nBox.nArea,50,200)        & ...
         isbetween(nBox.BoundingBox(3),7,25) & ...
         nBox.BoundingBox(4)>10;
    Box=round(nBox(idx).BoundingBox);

where isbetween is my "syntactic sugar" utility function

>> type isbetwween
function flg=isbetween(x,lo,hi)
% returns T for values within range of input
% SYNTAX:
%  [log] = iswithin(x,lo,hi)
%      returns T for x between lo and hi values, exclusive

flg= (x>lo) & (x<hi);

>>

I've several of these that serve simply to reduce clutter at the upper level; another is iswithin which is inclusive on both bounds.

NB: The statement

    nBox(posi).BoundingBox;         % Frame objects

is pointless; perhaps it was a debug line if removed the trailing semi-colon to echo the values during development.

ERRATUM

As Guillaume reminds me, since LHS is an array instead of reduced structure arrray must contatenate the results--

Box=round(cat(1,nBox(idx).BoundingBox));

[Recap]

The range of Length will vary from 3 to 8. So is there no way to preallocate the Box since it will vary everytime?

[dpb]

Well, yes, you can allocate Length elements but for only 8 max you'll not notice the difference unless this is buried deep in some nested construct.

But, as noted, throw away the loop and it'll all get allocated in "one swell foop" from the git-go and preallocating is then a non-issue.

Answer 2

[Guillaume]

As per dpb's answer, the loop and the if are completely unnecessary (and slow). Just use logical indexing on the whole array. As a result Box gets created in one go, when it receives the values and therefore there's no need to preallocate (as it would just slow down the code).

Now, dpb's answer is incorrect in the assignment to Box, but he's got the right idea. This will work:

areas = [nBox.Area];
bboxes = vertcat(nBox.BoundingBox);
Box = round(bboxes(areas > 50 & areas < 200 & bboxes(:, 4) > 10 ...
                   & bboxes(:, 3) > 7 & bboxes(:, 3) <  25, :));

Note that a comment explaining why the bounding box coordinates are rounded wouldn't go amiss. Assuming the coordinates come from regionprop, ceil would be more appropriate.

[dpb]

"dpb's answer is incorrect in the assignment to Box..."

Only in missing a closing parenthesis...corrected.

[Guillaume]

Only in missing a closing parenthesis

Not really. nBox is a structure array. So you've got to cat nbox.Area and nbox.BoundingBox.

[dpb]

Hmmm...oy, yeah...forgotted it being an array as the target instead of a reduced structure array. OK, so I fixed that, too... :)