matlab中的加速循环

Question

I'm very new in MATLAB (this is my first script). I wonder how may I speed up this loop, I don't know any toolbox or 'tricks' as I'm a newbie on it. I tried to code it with instinct, it works, but it is really long.

All are variables get with fread or integer manually entered, so this is basically simple math, but I have no clue on why is it so long (maybe nested loops?) and how to improve, as I am more familiar with Python and for example multiprocess .

Thanks a lot

X = 0;
Points = [0,0,0];

for i=1:nbLines

    for j=1:nbPositions-1
        if lDate(i)>posDate(j) && lDate(i)<=posDate(j+1)

            weight      = (lDate(i) - posDate(j))  / (posDate(j+1)- posDate(j));
            X    = posX(j)*(1-weight) + posX(j+1) * weight;
        end
    end
    
    if X ~= 0
        for j=1:nbScans

            Y = - distance(i,j) / tan(angle(i,j));
            Points = [Points;X, Y, distance(i,j)];

        end
    end
end

Answer 1

You have one issue with the given code. The blow line:

 Points = [Points; X, Y, distance(i,j)];

This will definitely slow up your code. You need to initialize this array to store the numbers. If you initialize it, you will find good difference in speed.

    X = 0;
Points = zeros([],3) ; 
Points(1,:) = [0,0,0];
count = 1 ; 


for i=1:nbLines
    
    for j=1:nbPositions-1
        if lDate(i)>posDate(j) && lDate(i)<=posDate(j+1)
            
            weight      = (lDate(i) - posDate(j))  / (posDate(j+1)- posDate(j));
            X    = posX(j)*(1-weight) + posX(j+1) * weight;  
        end
    end
    
    if X ~= 0
        for j=1:nbScans
            count = count+1 ; 
            Y = - distance(i,j) / tan(angle(i,j));
            Points(count,:) = [X, Y, distance(i,j)];
            
        end
    end
end

Note that, you code only saves the last value of X, is this what you want?

Answer 2

    X = 0;
Points = cell([],1) ; 
Points{1} = [0,0,0];
count = 1 ; 
for i=1:nbLines
    
    id = find(lDate(i)>posDate & lDate(i)<=posDate) ;
    if length(id) > 1 
        weight      = (lDate(i) - posDate(id(1)))  / (posDate(id(end))- posDate(id(1)));
        X    = posX(id(1))*(1-weight) + posX(id(end)) * weight;
    end
    
    
    if X ~= 0
        j=1:nbScans ; 
        count = count+1 ; 
        Y = - distance(i,j)./tan(angle(i,j));
        Points{count} = [repelem(X,size(Y,2),size(Y),1), Y, distance(i,j)'];                  
    end
end

Answer 3

try using parallelization- "parfor" instead of "for" that uses all available processors.

parfor i=1:nbLines
   rest of code here
end