在Matlab中使用'。*'操作时出现内存不足错误

Question

Below is the code I used in my m.file:

for idx = i_start:i_end
    CheckTemp = (timeTick > time_tr(idx)) .* (timeTick <= time_tr(idx));
    CheckTemp2 = find(CheckTemp);
    IdxS = min(CheckTemp2);
    IdxE = max(CheckTemp2);
    ......

the timeTick array is a double array about 100MB , and idx is from about 2000 to 4000 . Cause I check the m-file step by step, I find that before the loop, my memroy commit charge is: 817M/5422M . There is a lot of free space, right?

But my Matlab v6.5 told me that: Error in '.*', out of memory ... which makes me really confused.

Plus, I executed the code line by line before. And no error occurred. I really don't know why. I hope there is someone who can help...

If you need anymore information, please comment.

Answer 1

The MATLAB memory management page has some useful information such as the process memory limit on your system. In general, it is a good idea to run clear all before running a memory-intensive operation. Since, as Daniel pointed out, MATLAB's ability to allocate arrays is based on contiguous available memory, and clear does not guarantee garbage collection, it may be a good idea to restart MATLAB, espeially on older versions.

On a side note, I suspect that you will find that CheckTemp = (timeTick > time_tr(idx)) .* (timeTick <= time_tr(idx)); will always give you an array of logical zeros. .* Acts like binary & on arrays of logicals and you are guaranteeing that there will be no overlap by comparing > in one case and <= in the other. Perhaps you meant to use something like time_tr(idx-1) for one of the indices?

Answer 2

If timeTick and timeTr are sorted, this can be done:

1. Without creating additional arrays of the same huge size as timeTick
2. With a single pass through the two arrays timeTick and timeTr
3. In about 0.29 seconds in Matlab 2015b on a 150MB double array timeTick and 2000 entry array timeTr using the code below (on my computer).

The code would go as follows (note you must set time_tr(end+1) = inf for my code to work).

%This code requires last entry of time_tr to be inf, eg. time_tr(end+1)=inf;
n_tick = length(timeTick);
n_tr   = length(time_tr);

IdxS = NaN(n_tr, 1);
IdxE = NaN(n_tr, 1);

i_tick = 1;
i_tr   = 1;

window_start_i_tick = 1; 
window_end    = time_tr(1);

while(i_tick <= n_tick)
   t = timeTick(i_tick);
   if(t > window_end)
      IdxS(i_tr) = window_start_i_tick;
      IdxE(i_tr) = i_tick - 1;

      window_start_i_tick = i_tick;

      i_tr = i_tr + 1;
      while(t > time_tr(i_tr)) %take care of case that we skip past a window/windows
          i_tr = i_tr + 1;
      end

      window_end = time_tr(i_tr);        
   end    
   i_tick = i_tick + 1;
end
IdxS(i_tr) = window_start_i_tick;
IdxE(i_tr) = i_tick;

Answer 3

Reading the documentation for memory you will notice that the continuous available address space is the limit. Your Matlab version can not use your full memory, it is limited to a 32 bit addresses space.

I recommend to switch to a recent 64 bit version of matlab or to octave which is a open source clone.