Speeding up code by avoiding FOR, IF loops

Hi all, I have a question about how to speed up my code. I've been taught that for/if loops in matlab generally aren't great ideas, so I want to find a way to avoid them in the following code.

If you have suggestions on how to make these calculations outside of loops, I'd really appreciate it.

for i = 1:length(uniqueID)
  for j = 1:length(timeList)
    x = strmatch(uniqueID(i),idSeriesMonthly)
    y = blsDataMonthly(x,[1 4])'
    for k = 1:length(y)
      if blsDataSeries(1,j) == y(1,k)
        blsDataSeries(i + 1,j) = y(2,k)
      end
    end
  end
end

Answer 1

[Jan]

The rumor, that for-loops are slow in general concern the versions before 6.5 (July 2002). In modern Matlab versions the JIT can accelerate loops substantially.

In opposite to this strmatch is really slow. Better use strcmp or strncmp.

In addition the output to the command window consumes much time also. So append a semicolon after each assignment command.

You overwrite blsDataSeries(i + 1,j) repeatedly. So better start the innermost loop from the back and stop after the first match:

for k = length(y):-1:1
  if blsDataSeries(1,j) == y(1,k)
    blsDataSeries(i + 1,j) = y(2,k);
    break;  % Break out of the k-loop
  end
end

Avoid the transposition of y.

[EDITED]

nTime         = length(timeList);
blsDataSeries = zeros(length(uniqueID) + 1, nTime);  % pre-allocate!
q = zeros(1, nTime);

for i = 1:length(uniqueID)
  aID = uniqueID{i};
  x   = strncmp(aID, idSeriesMonthly, length(aID));
  if sum(x) > 2
    y1   = blsDataMonthly(x, 1);
    y4   = blsDataMonthly(x, 4);
    q(:) = NaN;  % Re-use existing memory
    for a = 1:length(y)
      y1a = y1(a);
      for b = nTime:-1:1
        if y1a == timeList(b)
          q(b) = y4(a);
          break;  % Break b-loop after the last occurrence
        end
      end
    end
    blsDataSeries(i + 1, :) = q;
  end
end

[Tim]

Hey thanks for your help! Unfortunately this specifically didn't really speed up the process much. The code I'll post below did help, basically I ended up creating a new row to insert into the matrix rather than doing multiple for loops to locate specific cells.

It's still pretty slow but it's running at a much better clip (maybe 30mins for the whole thing, rather than multiple days). I think there's probably a way to vectorize the first FOR loop, but I'm not sure about it yet.

for i = 1:length(uniqueID);

x = strmatch(uniqueID(i),idSeriesMonthly);

y = blsDataMonthly(x,[1 4])';

if length(x) > 2; %there is one solo entry which screws with the q row variable later, but i don't need it

q = NaN(1,length(timeList));

for a = 1:length(y);

for b = 1:length(timeList);

if y(1,a) == timeList(b);

q(b) = y(2,a);

end

end

end

blsDataSeries(i + 1,:) = q;

end

end

[Jan]

Is blsDataSeries pre-allocated properly? If not, this array grows in each iteration, which wastes a lot of time. See [EDITED]. It would be easier to help, if the types and sizes of the used variables are explained.