How to speed up this calculation and remove loop(s)?

Hello,

I have a table oFD that 10450x27 in size. I am trying to optimize this double loop to speed up the processing. For whatever reason, I'm drawing a blank how to optimize and am sure if can be computed more efficiently.

sPs = [3500:5000]';
CGE = zeros(size(sPs,1),1);

for row = 1:size(oFD,1)
    for m = 1:size(sPs,1)
        CG = my_function(sPs(m),oFD.PS(row),oFD.dTE(row),oFD.CIV(row));
        CG2(row,m) = oFD.COI(row)*sPs(m)^2*CG;
    end
end

Would someone please be able to help?

Thank you in advance!

[James Tursa]

This is going to depend largely on what my_function( ) does and whether it is or can be vectorized. You will need to show us the my_function( ) code.

[Image Analyst]

What is size(oFD,1) and size(sPs,1)? Unless they're tens of millions, your bottleneck may not be the loop iteration itself but something inside the loop. If the total number of iterations is like a few hundred thousand, you're only talking about microseconds for the looping overhead.

[Daulton_Benesuave]

Sorry for the late reply, had to attend to same family matters yesterday afternoon.

size(oFD,1) = 10500x27

size(sPs,1) = 1797x1

Answer 1

[Jan]

There is a very small potential for optimizing in the posted code:

for row = 1:size(oFD,1)
    c1 = oFD.PS(row);
    c2 = oFD.dTE(row);
    c3 = oFD.CIV(row);
    c4 = oFD.COI(row);
    for m = 1:size(sPs,1)
        CG = my_function(sPs(m), c1, c2, c3);
        CG2(row,m) = c4 * sPs(m)^2 * CG;
    end
end

I assume, the main time is spent in my_function. The profiler would reveal this.

If you post the code of my_function, further improvements are possible.

[Daulton_Benesuave]

Hi Jan,

Thanks for the very quick reply!

my_function is simply a modified version of the builtin Matlab function 'blsgamma'.

blsgamma has a one extra input 'rate' than can be assumed to be 0 for my_function.

Do this help you optimize further?

[Cel Kulasekaran]

blsgamma is already vectorized, so you do not need a for loop even if you have a matrix of current stock prices and corresponding excercise prices.

E.g. if

S0 = MxN matrix of stock prices

X = MxN matrix of excercise prices

etc.

just call this with G = blsgamma(S0, X, ....)

There is no need to iterate with a for loop, this should be faster.

[Image Analyst]

@Daulton_Benesuave, why didn't you answer my question above?

https://www.mathworks.com/matlabcentral/answers/1814635-how-to-speed-up-this-calculation-and-remove-loop-s#comment_2388780

[Cel Kulasekaran]

Don't think it matters, Daulton's custom function is blsgamma which is already vectorized...

[Jan]

I do not have the Financial Toolbox. So this is a dumb guess only:

for row = 1:size(oFD,1)
    c1 = oFD.PS(row);
    c2 = oFD.dTE(row);
    c3 = oFD.CIV(row);
    c4 = oFD.COI(row);
    CG = my_function(sPs, c1, c2, c3);
    CG2(row, :) = c4 * sPs.^2 .* CG;
end

So can you call my_function with a vector as 1st input?