[Octave-bug-tracker] [bug #45497] missing function: repelem

[Nick Jankowski]

Follow-up Comment #2, bug #45497 (project octave):

Looks good. A couple comments. I noticed you're using repmat even for the
simpler cases. Just went through a run length decoding exercise and what I
first found was discussion about using indexing tricks to avoid calling
repmat, can make a big performance improvement. 

E.g., the following two lines produce the same output:


>> v = 2;
>> element = [1 2 3 4 5 6];
>> repmat (element, 2, 1)(:)'
ans =

   1   1   2   2   3   3   4   4   5   5   6   6

>> element(ones(2,1),:)(:)'
ans =

   1   1   2   2   3   3   4   4   5   5   6   6


Additionally, doing a quick and (very) dirty timing test:

>> tic;for ii = 1:10000,repmat (element, v, 1)(:)';,endfor;toc
Elapsed time is 3.182 seconds.
>> tic;for ii = 1:10000,element(ones(v,1),:)(:)';,endfor;toc
Elapsed time is 0.207 seconds.


For v being a vector, I came upon the following :

Matlab array manipulation tips and tricks
http://home.online.no/~pjacklam/matlab/doc/index.html

In chapter 15.5.2 it had a section on run length encoding & decoding with 3
different scripts. while the author says the 1st one should be the slowest, my
timing of the three with octave (it was a touch more involved that the timing
test above) showed it to be the fastest.  with your variable naming the code
would be (haven't checked for accuracy with your code):

ii = cumsum(v); 
jj = zeros(1, ii(end));
jj(ii(1:end-1)+1) = 1; 
jj(1) = 1;
ret = element(cumsum(jj));
 

I've been toying with using the (ones()) indexing with multidimensional
arrays, and it works well. If I get a chance I'll look at what you have and
what repelem needs to cover.

    _______________________________________________________

Reply to this item at:

  <http://savannah.gnu.org/bugs/?45497>

_______________________________________________
  Message sent via/by Savannah
  http://savannah.gnu.org/