Put a matrix back to zeros (looking for an elegant solution)

Hello,

I have one matrix that is generated withe zeros:

A=zeros(5,4);

And, during the cycle matrix A get different values to make some math. I want that in the end of the cycle the matrix A come back to is original values so it can start it all over. Is there any elegant solution or should i just use the same code that used to initialize the matrix?

Answer 1

[Walter Roberson]

A(:) = 0;

[André Pacheco]

Exactly :D Thanks!

[Sean de Wolski]

This takes about 50% longer on my system:

[t1,t2] = deal(0);
A = zeros(2000);
for ii = 1:100
  tic
  A = zeros(2000);
  t1=t1+toc;
  tic
  A(:) = 0;
  t2=t2+toc;
end
[t1 t2]
%ans=    0.2117    0.3609

[Walter Roberson]

Sounds like something Mathworks should be optimizing then ;-)

What about

A = A - A;

?

[Sean de Wolski]

2% slower than A(:)=0.

We're aware of it Walter, don't worry!

[Jan]

Surprising. I cannot test this currently:

tic;
for ii = 1:100
  A = zeros(2000);
  A(1, 1:10) = rand(1, 10);
end
toc

tic;
A = zeros(2000);
for ii = 1:100
  A(:) = 0;
  A(1, 1:10) = rand(1, 10);
end
toc

Actually allocatiing the array should be more expensive than setting it to 0.

[Walter Roberson]

When I test in a development version, the timings cannot be reliably distinguished. A(:) = 0 did peak higher than zeros(2000), but the times strongly overlapped, and zeros(2000,2000) tended higher than zeros(2000) but again with overlap. And what trended as lowest was A = A - A

Tossing the possibilities into a function in a .m file might make a significant difference in this situation.

[Jan]

@Walter: As far as I understand, you do not observe a 50% difference on your system.

[Walter Roberson]

Now that I have turned off the processes using 7.1 of my 8 CPUs, I get much more reproducible results. Times still overlap, but in my test, A(:)=0 tends to be lower, and A=A-A; tends to be lower still. Maximum variability between all the possibilities was 0.305 (A-A) to 0.319 (zeros(2000,2000)). Not even close to 50%. (This for a simplified version without the rand())

[Walter Roberson]

R2012a on MacBook Pro, OS-X Lion, i7 CPUs, the A(:)=0 consistently tests slightly faster when I use Sean's code, not slower at all.

Sean, did you forget to turn off your disk defragger while you were testing? :)

Answer 2

[Sean de Wolski]

That's what I would do. zeros is the most elegant want to create zeros :)

[André Pacheco]

I thought there was something like A=clean; And it would clean the matrix :D Something like harry potter style :D

[Sean de Wolski]

You could write clean() to do this if you wanted, considerably less elegant though...