How can I improve my code by using logical indexing instead of "find" function.

TLDR; I want to improve the code given below;

SS(1) = 0;
SS(2) = 0.5;
SS(3) = 1;
SS(4) = 1.5;
SS(5) = 2;
SS(6) = 2.5;

for sigma=SS    
    for i=1:trialSize            
        estErr(i,find(SS==sigma))...                
        = Trilateration( ROI, gridSize, N_s, P_T, P_E, sigma, alpha_actual ...
                        , alpha_assumed, recSens ...
                        , dispON ...
                        , useTime, assignS, assignE);
    end            
    disp(['Shadow Spread ', num2str(SS(k)), ' is complete.']);
end

Detailed Background information; I have encountered a warning but couldn't think of a way to improving my code in the way suggested by the warning. The warning suggest that I use logical indexing instead of using find function. I am not allowed to use parfor instead of for loops here due to the way I use estErr in the code. Being able to parallelize my simulation will be a huge bonus for me. This way simulations take a long time probably due to the way I compute each estimated error obtained by each run of the function.

I am sure the loop given here and the totality of my code can be improved by adding basic coding practices. I learned MATLAB on my own, and I don't know how to "properly" write code. I wrote the code to implement localization techniques. I am trying to see how the technique fares in a simulation with changing Shadow Spread values. Sorry if my code makes you cringe. Any kind of improvement suggestion is welcome.

[Image Analyst]

Check this link out: Best ways to master MATLAB

[Kerem Karatas]

Thank you for the link. I will definitely try to make use of the information provided in that link.

Answer 1

[Ahmet Cecen]

There are many many ways to improve that code, but the specific error you are talking about refers to:

 estErr(i,find(SS==sigma))

and it is telling you to use:

 estErr(i,SS==sigma)

instead.

[Kerem Karatas]

The outputs of find function and the logical indexing is given below;

>> find(SS==sigma)
ans =
     3
>> SS==sigma
ans =
     0     0     1     0     0     0

I am checking my code and it looks like its working as intended. How does estErr(i,SS==sigma) find the right index?

[Kerem Karatas]

Actually looking over again my code is not working as intended. I want the code to run trialSize x length(SS) simulations and put the results into estErr 2D array. Using SS==sigma instead of find(SS==sigma) always writes to the first column of estErr instead of incrementing when the SS value used is increased.

I am not getting an error, it is just a warning and using find(SS=sigma) code works as I intend it to work. I just want to know what is the right way of writing this kind of code.

[Ahmet Cecen]

Interesting, I get identical results. Any ways, it doesn't matter I would take up image analyst on that link, this is a very very roundabout way of indexing to begin with. Running it as it is if you get the answer you want is okay, unless you run this with a 1x1000000 SS vector later, you will not see the effects of that performance drop the error is referring to.

 A=rand(5);
 sigma=3;

 A=

    0.8147    0.0975    0.1576    0.1419    0.6557
    0.9058    0.2785    0.9706    0.4218    0.0357
    0.1270    0.5469    0.9572    0.9157    0.8491
    0.9134    0.9575    0.4854    0.7922    0.9340
    0.6324    0.9649    0.8003    0.9595    0.6787

 find(SS==sigma)

 ans =

     3

 A(2,find(SS==sigma))

 ans =

     0.9706

 A(2,SS==sigma)

 ans =

     0.9706

[Kerem Karatas]

It works like how you explained. I had the same conclusion before I made a mistake by using "=" instead of "==" which confused me even more. Now reading and learning more about logical indexing I have a better understanding on addressing an element of an array.

Thank you for detailed explanation.

Answer 2

[Edric Ellis]

I would try something along these lines:

parfor sidx = 1:numel(SS)
  sigma = SS(sidx);
  for i = 1:trialSize
    estErr(i, sidx) = ...
  end
end

The trick here is to make the loop variable operate over the number of elements of SS, and then deduce sigma from that.

[Kerem Karatas]

Thank you, when I adjusted my code it allowed parfor to be used instead of for. From what I read I gathered it is not possible to have two parallel for loops in a nested manner. That makes sense to me. But also makes me think...

Function calls are done in the inner loop and trialSize is much larger than numel(SS) therefore instead of making the outer loop parfor I changed the inner loop to parfor. Calls of the same function should be independent if I am not mistaken.

Thanks again for helping me understand how to make my code work parallel.

[Edric Ellis]

It's usually best to make the outer loop the parfor, but if you don't have enough parallelism there, then it makes sense to make the inner loop the parfor. Basically it comes down to balancing the amount of time wasted when there isn't enough parallelism (when some workers are idle) vs. the overhead of launching a parfor loop.