Filter löschen
Filter löschen

performance of diff on a logical array

4 Ansichten (letzte 30 Tage)
Tom DeLonge
Tom DeLonge am 27 Feb. 2018
Kommentiert: Tom DeLonge am 9 Mär. 2018
I have a 5000x5000 logical array. I want to find the indices where the logical array has a rising edge, e.g. whenever the array goes from 0 to 1 within a row. Until now I did this as follows:
rising_edges = diff(A) == 1;
However, I notice that this is slow. I am pretty certain that this is slow because diff(A) has to allocate new memory for a 5000x5000 double array.
I am wondering: is there a more elegant and faster way of doing this? I'd also be happy with just the subscripts instead of a full logical index array.

Akzeptierte Antwort

Walter Roberson
Walter Roberson am 9 Mär. 2018
rising_edges = A(:,2:end)) & ~A(:,1:end-1);
this operates on the entire array at the same time. You can use the two-output version of find() on the result, prioritized by row instead of by column. You can put the two outputs together and sortrows() or there are various other ways like findgroups()
  4 Kommentare
Tom DeLonge
Tom DeLonge am 9 Mär. 2018
oh this is a nice contribution. Indeed, something I've been looking for a while. Let's see whether it is 64-bit compatible...
Tom DeLonge
Tom DeLonge am 9 Mär. 2018
Walter, I checked the contribution. The results are indeed interesting. It turns out, that using sharedchild I get a speedup by a factor of 2. I was quite surprised, I would have expected a 5-fold speedup (as indicated by my stats above). Turns out, that now the actual binary operation takes longer than before (nearly 4 times). The sharedchild-operation however is incredibly fast.
I have no idea what is going on behind the scenes here...

Melden Sie sich an, um zu kommentieren.

Weitere Antworten (1)

lokender Rawat
lokender Rawat am 9 Mär. 2018
You can simply declare an array "mat" and do the following, you do not need to create another new double array here. It will find the index from the logical array for each row and put it in "index" variable.
[matRow,matCol]=size(mat)
for i=1:matRow
index=find(mat(i,2:end)-mat(i,1:end-1)==1)
end
  2 Kommentare
Walter Roberson
Walter Roberson am 9 Mär. 2018
This will overwrite the entire variable index each time through the loop.
Tom DeLonge
Tom DeLonge am 9 Mär. 2018
yes, Robert, thanks for that comment. I was able to increase the speed by a factor of 2 when not using diff, but rather having
A(1:end-1) > A(2:end)
still not really fast, since two logical arrays seem to be allocated here...

Melden Sie sich an, um zu kommentieren.

Kategorien

Mehr zu Loops and Conditional Statements finden Sie in Help Center und File Exchange

Community Treasure Hunt

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

Start Hunting!

Translated by