elimination of conscutive regions (generalization: ones with zeros between)

1 vue (au cours des 30 derniers jours)

I need effectively eliminate (by zeroing) the consecutive "1's" between "-1's" and start/end of column at each column of matrix A, which now can be separated by any number of zeroes. The number of consecutive "1's" between "-1's" and start/end of column is > N. This is a non-trivial generalization of my previous Question.

Again, typical size(A) = [100000,1000].

See example:

A =
     1    -1     0
     0     1     1
     0     1     1
     1     1     0
     0     0     1
     1    -1     0
    -1     1     1
    -1     0    -1
     1     1     1
     0     1    -1

For N = 2 the expected result is

Aclean =
     0    -1     0
     0     0     0
     0     0     0
     0     0     0
     0     0     0
     0    -1     0
    -1     0     0
    -1     0    -1
     1     0     1
     0     0    -1

For N = 3 the expected result is

Aclean =
    1    -1     0
    0     1     0
    0     1     0
    1     1     0
    0     0     0
    1    -1     0
   -1     1     0
   -1     0    -1
    1     1     1
    0     1    -1
  4 commentaires
Matt J
Matt J le 1 Oct 2018
Modifié(e) : Matt J le 1 Oct 2018
which now can be separated by any number of zeroes
This is the definition of non-consecutive. Why, then, are you saying the eliminated 1's are to be consecutive?
Michal
Michal le 1 Oct 2018
Well OK, any sequence of "1's" and "0's" separated by start/end and "-1's".

Connectez-vous pour commenter.

Réponse acceptée

Bruno Luong
Bruno Luong le 1 Oct 2018
A = [1 -1 0;
0 1 1;
0 1 1;
1 1 0;
0 0 1;
1 -1 0;
-1 1 1;
-1 0 -1;
1 1 1;
0 1 -1];
N = 3;
[m,n] = size(A);
Aclean = A;
for j=1:n
Aj = [-1; A(:,j); -1];
i = find(Aj == -1);
c = histc(find(Aj==1),i);
b = c <= N;
im = i(b);
ip = i([false; b(1:end-1)]);
a = accumarray(im,1,[m+2,1])-accumarray(ip,1,[m+2,1]);
mask = cumsum(a);
mask(i) = 1;
Aclean(:,j) = Aclean(:,j).*mask(2:end-1);
end
Aclean
  3 commentaires
Michal
Michal le 2 Oct 2018
Is there any serious reason to still use old "histc" instead "histcounts"? Both functions has same performance and "histc" will be removed in future releases.
Bruno Luong
Bruno Luong le 2 Oct 2018
No, I'm just used to HISTC.

Connectez-vous pour commenter.

Plus de réponses (2)

Michal
Michal le 1 Oct 2018
Modifié(e) : Michal le 1 Oct 2018
I am still looking for better (faster) solution. Any idea how to improve so far best solution:
A = [1 -1 0;
0 1 1;
0 1 1;
1 1 0;
0 0 1;
1 -1 0;
-1 1 1;
-1 0 -1;
1 1 1;
0 1 -1];
N = 3;
sep = A==-1;
sep(1,:) = true;
idx = cumsum(sep(:));
sep(1,:) = A(1,:)==-1;
num = accumarray(idx, A(:)==1);
iff = num <= N;
Aclean = reshape(sep(:)|iff(idx), size(A)) .* A;
Aclean
Big test matrix:
A = double(rand(100000,1000)>.4) - double(rand(100000,1000)>.65);
Bruno's code (N = 5): Elapsed time is 4.848747 seconds.
My code (N = 5): Elapsed time is 3.257089 seconds.
  13 commentaires
Michal
Michal le 2 Oct 2018
Yes … nice simplification!
Bruno Luong
Bruno Luong le 2 Oct 2018
And what about
mask = Aclean==A
?

Connectez-vous pour commenter.


Bruno Luong
Bruno Luong le 2 Oct 2018
That's true, somehow it's a 1D scanning problem.
I think we are close to the limit of MATLAB can do, if faster speed is still needed, then one should go to MEX programming route instead of torturing MATLAB to squeeze out the last once of speed.

Produits


Version

R2018a

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by