How to speed up simple sorting

Hello, I got two variables (Units and time points) that needs to me sorted with respect to each other. Meaning, I want to sort all the time points to its specific Unit. With large numbers of Units and many time points this sorting takes forever. Is there a way to speed this up?

What I got so far works, but it is rather slow:

%Prepare example RAW data
Unit=int32(randi(60,5000000,1)); %60 Units in an order, that matches the timepoints
timepoints=randi(600,5000000,1); %5*10^6 Timepoints (from a range of 600ms) in an order, that corresponds to the according unit
%Prepare the variable where the data should be sorted
for i=1:60
    Unit_sort{i,1}=i;
    Unit_sort{i,2}=[];
end

%Sort "time points" to the according "Unit"
for i=1:length(timepoints)
    index = find([Unit_sort{:,1}] == Unit(i));
    Unit_sort{index,2}=[Unit_sort{index,2} timepoints(i)];
end

[Jan]

Typo: "Units(i)" must be "Unit(i)". But after fixing this, the code fails, because find([Unit_sort{:,1}] == Units(i)) must be empty from the beginning.

Please post some code, which performs, what you want to achieve.

[Stephen23]

Start by reading this:

https://www.mathworks.com/help/matlab/matlab_prog/techniques-for-improving-performance.html

Answer 1

[Jan]

2 general rules:

• Pre-allocate the output. The iterative growing of arrays is a Don't!
• Prefer to run the loop over the smaller array.

uUnit     = unique(Unit);
Unit_sort = cell(1, length(uUnit));
for k = 1:length(uUnit)
   Unit_sort{k} = timepoints(Unit == uUnit(k));
end

Perhaps this does what you want - I'm not sure because your posted code does not run and I cannot guess, what the wanted result is. But if it runs, read the docs of splitapply and accumarray.

[Felix Ludwig]

Thank you for your help, I corrected the code I provided - now it works (In the old slow way).

I will try to apply your suggestions now and see, if it speeds up the procedure ...

[Felix Ludwig]

ok, I tried your code and indeed its waaaay faster. Thank you! Note to myself: Always run the loop over the smaller array!

Answer 2

[Steven Lord]

I think what you're trying to do is sortrows.

% Sample data
A = randi(10, [20 2]);

% Sort by column 1, use column 2 to break ties
B = sortrows(A);

% Sort by column 2, use column 1 to break ties
C = sortrows(A, [2 1]);

% Show all three side-by-side
T = table(A, B, C)

[Felix Ludwig]

Thank you for your input, It's not exactly what I meant. I want to have for example 60 unique Units and sort a high number of timepoints to the unit they belong to. I updated the provided code above. Now it works - in the original slow way. But I will try to apply the sortrow procedure to my problem.