Finding the number of rows to the next row containing a 1
4 visualizzazioni (ultimi 30 giorni)
Mostra commenti meno recenti
Hi
I have a column vector of 1s and 0s and I want to find find the number of rows to the next row containing a 1. For example:
A = [0 0 0 1 0 1 1 1 0 0 0 0 0 1]';
I would like the code to return
B = [3 2 1 0 1 0 0 0 5 4 3 2 1 0]';
Is there a vectorized way that this can be done?
Thanks in advance
10 Commenti
Bruno Luong
il 28 Ago 2020
Very convincing Stephen. In my laptop it's even more obvious
t Rik = 0.680010 [s]
t Bruno = 0.378046 [s]
t Stephen = 0.107799 [s]
In your for-loop code I would cast B initialization in double (in case A is logical)
B = double(A);
or
B = zeros(size(A));
Bruno Luong
il 28 Ago 2020
Modificato: Bruno Luong
il 28 Ago 2020
As I'm slightly surprised by the performance of the for-loop (I would expext it's good but not THAT good), I then try to see how far it from a MEX implementation. And I'm stunned, it's almost as fast (even faster for smaller array).
t Rik = 0.651531 [s]
t Bruno = 0.379362 [s]
t Stephen = 0.104442 [s]
t MEX = 0.073168 [s]
The code (benchmark + Cmex) is in the attacheh file for those who wants to play with.
I must congratulat TMW for improving the for-loop performance over many years.
Risposta accettata
Rik
il 27 Ago 2020
Modificato: Rik
il 27 Ago 2020
It took some time, but here is a solution that should also work for large matrices.
clc,clear
format compact
A = [0 0 0 1 0 1 1 1 0 0 0 0 0 1]';
% 3 2 1 0 1 0 0 0 5 4 3 2 1 0
B=A;
pad=B(end)~=1;
if pad,B(end+1)=1;end %this method requires the last position to be a 1
B=flipud(B);
C=zeros(size(B));
C(B==1)=[0;diff(find(B))];
C=ones(size(B))-C;
out=cumsum(C)-1;
out=flipud(out);
if pad,out(end)=[];end
%only for display:
[A out]
4 Commenti
Rik
il 27 Ago 2020
If that is indeed a problem that should be easy to fix. I chose to write a comment instead of changing that behavior, but maybe I should have made it more explicit. Thank you for drawing more attention to that point.
Più risposte (3)
Binbin Qi
il 27 Ago 2020
A = [0 0 0 1 0 1 1 1 0 0 0 0 0 1]';
C = find(A);
D = (1:length(A)) - C;
D(D>0) = D(D>0) + inf';
min(abs(D))'
ans =
3
2
1
0
1
0
0
0
5
4
3
2
1
0
6 Commenti
Bruno Luong
il 27 Ago 2020
Modificato: Bruno Luong
il 27 Ago 2020
As much as I love vectorization, this problem is a typical case where the for-loop method is easier, faster, more readable.
This code is ready for 2D array, it works along the first dimension independently.
A=rand(30000,1000)>0.7;
tic
Al=logical(A);
B=zeros(size(A));
b=B(1,:);
for k=size(B,1):-1:1
b = b+1;
b(Al(k,:))=0;
B(k,:)=b;
end
toc
Bruno Luong
il 28 Ago 2020
Modificato: Bruno Luong
il 28 Ago 2020
Now I just discover CUMSUM has direction option, this is based on Rik's cumsum method, but avoid the double flipping.
B = ones(size(A));
i1 = find(A);
B(i1) = 1-diff([i1;size(A,1)+1]);
B = cumsum(B,'reverse');
1 Commento
Rik
il 28 Ago 2020
Cool, I didn't remember that was an option. Turns out that is already an option as far back as R2015a. The release notes no longer allow you to look back further than R2015a, even if you try modifying the url. All I know is that it isn't possible in R2011a.
Vedere anche
Categorie
Scopri di più su Matrix Indexing in Help Center e File Exchange
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!