Remove for loop and reduce time

I want to reduce the running time of my code. I found that most of the time in my code is spent in this loop, as this loop is invoked multiple times in my code. Is there anyway to reduce time?? Code explanation: ac1 is an acceleration matrix having values ranging from -10 to 10. I am using discretize function to group different acceleration values into different bins, represented by edges_a. Now, I have to find how many values are there in each bin and store those values in the fourth column of acc_dis matrix.

Here is the code:

if true
  edges_a=min(ac1):1.25:max(ac1);
  grp_ac=discretize(ac1,edges_a);

    for i=1:length(edges_a)
    acc_dis(i,4)=sum(grp_ac==i);
    end
  end

[Nithin Banka]

So you want to replace 4th column of 'acc_dis' with the number of times integer 'i' appears in 'grp_ac'?

[Nagesh A P]

@Nithin, yes

[Nithin Banka]

Then, what I though of is, instead of calling sum() in every loop iteration, since 'i' cannot be beyond the length 17. So we define a temporary array with the length of 'i' and then run a loop through 'grp_ac' and increment the temporary array element according to the number. That way we just need to run through all the elements in grp_ac once to get the required answer. I hope this will take less time.

I know this is not so clear. Can you tell me the size of 'grp_ac'? I will try reproducing what I told you and post you the code if I get good results.

[Nagesh A P]

grp_ac is an 18000x1 array. It's not fixed, it will change in each iteration. But always between 17500 to 18500. But that won't matter to you. U can use even a 100x1 array as grp_ac for example I guess.

[Jan]

@Nagesh A P: Please post a relevant set for the input data.

Answer 1

[Guillaume]

Points for using discretize instead of writing your own loop for that. The thing is, what you really want is the histogram and there's also a function for that: histcounts:

edges_a = min(ac1):1.25:max(ac1);
h = histcounts(ac1, edges_a);
acc_dis(1:numel(h), 4) = h;

Note: I don't know how many rows acc_dis has. Hopefully at least enough. If it already has the correct number of rows, that last line can be simplified to:

acc_dis(:, 4) = h;

Note 2: If you still need grp_ac, then:

[h, ~, grp_ac] = histcounts(ac1, edges_a);

[Nagesh A P]

That was what I was looking for. Thanks.

Answer 2

[Nithin Banka]

if true
    edges_a=min(ac1):1.25:max(ac1);
    grp_ac=discretize(ac1,edges_a);
    len = length(edges_a);
    auxArr = zeros(len, 1));
    len1 = length(grp_ac);
    for i=1:len1
        if(grp_ac(i)<=len&grp_ac(i)>=1)
            auxArr(grp_ac(i)) = auxArr(grp_ac(i)) + 1;
        end
    end
    acc_dis(1:len, 4) = auxArr;
end

I think this should help.

[Nagesh A P]

Ur code will work. But I think this will take more time than my code. Ur code also uses an loop. I'm working with really big data and there will be thousands of iterations in that loop.So I'm losing a lot of time in loops. So what I've asked is for a code that saves my time, if possible.

[Nithin Banka]

The code you used runs through grp_ac in every iteration in the 'grp_ac==i' command. That will make it kind of for loop inside a for loop.

[Nagesh A P]

Ya. i just checked times. Ur code takes only about half as mine. thanks. But can u optimize it more??